drivers/md/md.c

   1 /*
   2    md.c : Multiple Devices driver for Linux
   3           Copyright (C) 1998, 1999, 2000 Ingo Molnar
   4
   5      completely rewritten, based on the MD driver code from Marc Zyngier
   6
   7    Changes:
   8
   9    - RAID-1/RAID-5 extensions by Miguel de Icaza, Gadi Oxman, Ingo Molnar
  10    - RAID-6 extensions by H. Peter Anvin <hpa@zytor.com>
  11    - boot support for linear and striped mode by Harald Hoyer <HarryH@Royal.Net>
  12    - kerneld support by Boris Tobotras <boris@xtalk.msk.su>
  13    - kmod support by: Cyrus Durgin
  14    - RAID0 bugfixes: Mark Anthony Lisher <markal@iname.com>
  15    - Devfs support by Richard Gooch <rgooch@atnf.csiro.au>
  16
  17    - lots of fixes and improvements to the RAID1/RAID5 and generic
  18      RAID code (such as request based resynchronization):
  19
  20      Neil Brown <neilb@cse.unsw.edu.au>.
  21
  22    - persistent bitmap code
  23      Copyright (C) 2003-2004, Paul Clements, SteelEye Technology, Inc.
  24
  25    This program is free software; you can redistribute it and/or modify
  26    it under the terms of the GNU General Public License as published by
  27    the Free Software Foundation; either version 2, or (at your option)
  28    any later version.
  29
  30    You should have received a copy of the GNU General Public License
  31    (for example /usr/src/linux/COPYING); if not, write to the Free
  32    Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  33 */
  34
  35 #include <linux/module.h>
  36 #include <linux/config.h>
  37 #include <linux/kthread.h>
  38 #include <linux/linkage.h>
  39 #include <linux/raid/md.h>
  40 #include <linux/raid/bitmap.h>
  41 #include <linux/sysctl.h>
  42 #include <linux/devfs_fs_kernel.h>
  43 #include <linux/buffer_head.h> /* for invalidate_bdev */
  44 #include <linux/suspend.h>
  45 #include <linux/poll.h>
  46
  47 #include <linux/init.h>
  48
  49 #include <linux/file.h>
  50
  51 #ifdef CONFIG_KMOD
  52 #include <linux/kmod.h>
  53 #endif
  54
  55 #include <asm/unaligned.h>
  56
  57 #define MAJOR_NR MD_MAJOR
  58 #define MD_DRIVER
  59
  60 /* 63 partitions with the alternate major number (mdp) */
  61 #define MdpMinorShift 6
  62
  63 #define DEBUG 0
  64 #define dprintk(x...) ((void)(DEBUG && printk(x)))
  65
  66
  67 #ifndef MODULE
  68 static void autostart_arrays (int part);
  69 #endif
  70
  71 static LIST_HEAD(pers_list);
  72 static DEFINE_SPINLOCK(pers_lock);
  73
  74 /*
  75  * Current RAID-1,4,5 parallel reconstruction 'guaranteed speed limit'
  76  * is 1000 KB/sec, so the extra system load does not show up that much.
  77  * Increase it if you want to have more _guaranteed_ speed. Note that
  78  * the RAID driver will use the maximum available bandwidth if the IO
  79  * subsystem is idle. There is also an 'absolute maximum' reconstruction
  80  * speed limit - in case reconstruction slows down your system despite
  81  * idle IO detection.
  82  *
  83  * you can change it via /proc/sys/dev/raid/speed_limit_min and _max.
  84  * or /sys/block/mdX/md/sync_speed_{min,max}
  85  */
  86
  87 static int sysctl_speed_limit_min = 1000;
  88 static int sysctl_speed_limit_max = 200000;
  89 static inline int speed_min(mddev_t *mddev)
  90 {
  91         return mddev->sync_speed_min ?
  92                 mddev->sync_speed_min : sysctl_speed_limit_min;
  93 }
  94
  95 static inline int speed_max(mddev_t *mddev)
  96 {
  97         return mddev->sync_speed_max ?
  98                 mddev->sync_speed_max : sysctl_speed_limit_max;
  99 }
 100
 101 static struct ctl_table_header *raid_table_header;
 102
 103 static ctl_table raid_table[] = {
 104         {
 105                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MIN,
 106                 .procname       = "speed_limit_min",
 107                 .data           = &sysctl_speed_limit_min,
 108                 .maxlen         = sizeof(int),
 109                 .mode           = 0644,
 110                 .proc_handler   = &proc_dointvec,
 111         },
 112         {
 113                 .ctl_name       = DEV_RAID_SPEED_LIMIT_MAX,
 114                 .procname       = "speed_limit_max",
 115                 .data           = &sysctl_speed_limit_max,
 116                 .maxlen         = sizeof(int),
 117                 .mode           = 0644,
 118                 .proc_handler   = &proc_dointvec,
 119         },
 120         { .ctl_name = 0 }
 121 };
 122
 123 static ctl_table raid_dir_table[] = {
 124         {
 125                 .ctl_name       = DEV_RAID,
 126                 .procname       = "raid",
 127                 .maxlen         = 0,
 128                 .mode           = 0555,
 129                 .child          = raid_table,
 130         },
 131         { .ctl_name = 0 }
 132 };
 133
 134 static ctl_table raid_root_table[] = {
 135         {
 136                 .ctl_name       = CTL_DEV,
 137                 .procname       = "dev",
 138                 .maxlen         = 0,
 139                 .mode           = 0555,
 140                 .child          = raid_dir_table,
 141         },
 142         { .ctl_name = 0 }
 143 };
 144
 145 static struct block_device_operations md_fops;
 146
 147 static int start_readonly;
 148
 149 /*
 150  * We have a system wide 'event count' that is incremented
 151  * on any 'interesting' event, and readers of /proc/mdstat
 152  * can use 'poll' or 'select' to find out when the event
 153  * count increases.
 154  *
 155  * Events are:
 156  *  start array, stop array, error, add device, remove device,
 157  *  start build, activate spare
 158  */
 159 static DECLARE_WAIT_QUEUE_HEAD(md_event_waiters);
 160 static atomic_t md_event_count;
 161 void md_new_event(mddev_t *mddev)
 162 {
 163         atomic_inc(&md_event_count);
 164         wake_up(&md_event_waiters);
 165 }
 166 EXPORT_SYMBOL_GPL(md_new_event);
 167
 168 /*
 169  * Enables to iterate over all existing md arrays
 170  * all_mddevs_lock protects this list.
 171  */
 172 static LIST_HEAD(all_mddevs);
 173 static DEFINE_SPINLOCK(all_mddevs_lock);
 174
 175
 176 /*
 177  * iterates through all used mddevs in the system.
 178  * We take care to grab the all_mddevs_lock whenever navigating
 179  * the list, and to always hold a refcount when unlocked.
 180  * Any code which breaks out of this loop while own
 181  * a reference to the current mddev and must mddev_put it.
 182  */
 183 #define ITERATE_MDDEV(mddev,tmp)                                        \
 184                                                                         \
 185         for (({ spin_lock(&all_mddevs_lock);                            \
 186                 tmp = all_mddevs.next;                                  \
 187                 mddev = NULL;});                                        \
 188              ({ if (tmp != &all_mddevs)                                 \
 189                         mddev_get(list_entry(tmp, mddev_t, all_mddevs));\
 190                 spin_unlock(&all_mddevs_lock);                          \
 191                 if (mddev) mddev_put(mddev);                            \
 192                 mddev = list_entry(tmp, mddev_t, all_mddevs);           \
 193                 tmp != &all_mddevs;});                                  \
 194              ({ spin_lock(&all_mddevs_lock);                            \
 195                 tmp = tmp->next;})                                      \
 196                 )
 197
 198
 199 static int md_fail_request (request_queue_t *q, struct bio *bio)
 200 {
 201         bio_io_error(bio, bio->bi_size);
 202         return 0;
 203 }
 204
 205 static inline mddev_t *mddev_get(mddev_t *mddev)
 206 {
 207         atomic_inc(&mddev->active);
 208         return mddev;
 209 }
 210
 211 static void mddev_put(mddev_t *mddev)
 212 {
 213         if (!atomic_dec_and_lock(&mddev->active, &all_mddevs_lock))
 214                 return;
 215         if (!mddev->raid_disks && list_empty(&mddev->disks)) {
 216                 list_del(&mddev->all_mddevs);
 217                 /* that blocks */
 218                 blk_cleanup_queue(mddev->queue);
 219                 /* that also blocks */
 220                 kobject_unregister(&mddev->kobj);
 221                 /* result blows... */
 222         }
 223         spin_unlock(&all_mddevs_lock);
 224 }
 225
 226 static mddev_t * mddev_find(dev_t unit)
 227 {
 228         mddev_t *mddev, *new = NULL;
 229
 230  retry:
 231         spin_lock(&all_mddevs_lock);
 232         list_for_each_entry(mddev, &all_mddevs, all_mddevs)
 233                 if (mddev->unit == unit) {
 234                         mddev_get(mddev);
 235                         spin_unlock(&all_mddevs_lock);
 236                         kfree(new);
 237                         return mddev;
 238                 }
 239
 240         if (new) {
 241                 list_add(&new->all_mddevs, &all_mddevs);
 242                 spin_unlock(&all_mddevs_lock);
 243                 return new;
 244         }
 245         spin_unlock(&all_mddevs_lock);
 246
 247         new = kzalloc(sizeof(*new), GFP_KERNEL);
 248         if (!new)
 249                 return NULL;
 250
 251         new->unit = unit;
 252         if (MAJOR(unit) == MD_MAJOR)
 253                 new->md_minor = MINOR(unit);
 254         else
 255                 new->md_minor = MINOR(unit) >> MdpMinorShift;
 256
 257         init_MUTEX(&new->reconfig_sem);
 258         INIT_LIST_HEAD(&new->disks);
 259         INIT_LIST_HEAD(&new->all_mddevs);
 260         init_timer(&new->safemode_timer);
 261         atomic_set(&new->active, 1);
 262         spin_lock_init(&new->write_lock);
 263         init_waitqueue_head(&new->sb_wait);
 264
 265         new->queue = blk_alloc_queue(GFP_KERNEL);
 266         if (!new->queue) {
 267                 kfree(new);
 268                 return NULL;
 269         }
 270         set_bit(QUEUE_FLAG_CLUSTER, &new->queue->queue_flags);
 271
 272         blk_queue_make_request(new->queue, md_fail_request);
 273
 274         goto retry;
 275 }
 276
 277 static inline int mddev_lock(mddev_t * mddev)
 278 {
 279         return down_interruptible(&mddev->reconfig_sem);
 280 }
 281
 282 static inline void mddev_lock_uninterruptible(mddev_t * mddev)
 283 {
 284         down(&mddev->reconfig_sem);
 285 }
 286
 287 static inline int mddev_trylock(mddev_t * mddev)
 288 {
 289         return down_trylock(&mddev->reconfig_sem);
 290 }
 291
 292 static inline void mddev_unlock(mddev_t * mddev)
 293 {
 294         up(&mddev->reconfig_sem);
 295
 296         md_wakeup_thread(mddev->thread);
 297 }
 298
 299 static mdk_rdev_t * find_rdev_nr(mddev_t *mddev, int nr)
 300 {
 301         mdk_rdev_t * rdev;
 302         struct list_head *tmp;
 303
 304         ITERATE_RDEV(mddev,rdev,tmp) {
 305                 if (rdev->desc_nr == nr)
 306                         return rdev;
 307         }
 308         return NULL;
 309 }
 310
 311 static mdk_rdev_t * find_rdev(mddev_t * mddev, dev_t dev)
 312 {
 313         struct list_head *tmp;
 314         mdk_rdev_t *rdev;
 315
 316         ITERATE_RDEV(mddev,rdev,tmp) {
 317                 if (rdev->bdev->bd_dev == dev)
 318                         return rdev;
 319         }
 320         return NULL;
 321 }
 322
 323 static struct mdk_personality *find_pers(int level, char *clevel)
 324 {
 325         struct mdk_personality *pers;
 326         list_for_each_entry(pers, &pers_list, list) {
 327                 if (level != LEVEL_NONE && pers->level == level)
 328                         return pers;
 329                 if (strcmp(pers->name, clevel)==0)
 330                         return pers;
 331         }
 332         return NULL;
 333 }
 334
 335 static inline sector_t calc_dev_sboffset(struct block_device *bdev)
 336 {
 337         sector_t size = bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
 338         return MD_NEW_SIZE_BLOCKS(size);
 339 }
 340
 341 static sector_t calc_dev_size(mdk_rdev_t *rdev, unsigned chunk_size)
 342 {
 343         sector_t size;
 344
 345         size = rdev->sb_offset;
 346
 347         if (chunk_size)
 348                 size &= ~((sector_t)chunk_size/1024 - 1);
 349         return size;
 350 }
 351
 352 static int alloc_disk_sb(mdk_rdev_t * rdev)
 353 {
 354         if (rdev->sb_page)
 355                 MD_BUG();
 356
 357         rdev->sb_page = alloc_page(GFP_KERNEL);
 358         if (!rdev->sb_page) {
 359                 printk(KERN_ALERT "md: out of memory.\n");
 360                 return -EINVAL;
 361         }
 362
 363         return 0;
 364 }
 365
 366 static void free_disk_sb(mdk_rdev_t * rdev)
 367 {
 368         if (rdev->sb_page) {
 369                 put_page(rdev->sb_page);
 370                 rdev->sb_loaded = 0;
 371                 rdev->sb_page = NULL;
 372                 rdev->sb_offset = 0;
 373                 rdev->size = 0;
 374         }
 375 }
 376
 377
 378 static int super_written(struct bio *bio, unsigned int bytes_done, int error)
 379 {
 380         mdk_rdev_t *rdev = bio->bi_private;
 381         mddev_t *mddev = rdev->mddev;
 382         if (bio->bi_size)
 383                 return 1;
 384
 385         if (error || !test_bit(BIO_UPTODATE, &bio->bi_flags))
 386                 md_error(mddev, rdev);
 387
 388         if (atomic_dec_and_test(&mddev->pending_writes))
 389                 wake_up(&mddev->sb_wait);
 390         bio_put(bio);
 391         return 0;
 392 }
 393
 394 static int super_written_barrier(struct bio *bio, unsigned int bytes_done, int error)
 395 {
 396         struct bio *bio2 = bio->bi_private;
 397         mdk_rdev_t *rdev = bio2->bi_private;
 398         mddev_t *mddev = rdev->mddev;
 399         if (bio->bi_size)
 400                 return 1;
 401
 402         if (!test_bit(BIO_UPTODATE, &bio->bi_flags) &&
 403             error == -EOPNOTSUPP) {
 404                 unsigned long flags;
 405                 /* barriers don't appear to be supported :-( */
 406                 set_bit(BarriersNotsupp, &rdev->flags);
 407                 mddev->barriers_work = 0;
 408                 spin_lock_irqsave(&mddev->write_lock, flags);
 409                 bio2->bi_next = mddev->biolist;
 410                 mddev->biolist = bio2;
 411                 spin_unlock_irqrestore(&mddev->write_lock, flags);
 412                 wake_up(&mddev->sb_wait);
 413                 bio_put(bio);
 414                 return 0;
 415         }
 416         bio_put(bio2);
 417         bio->bi_private = rdev;
 418         return super_written(bio, bytes_done, error);
 419 }
 420
 421 void md_super_write(mddev_t *mddev, mdk_rdev_t *rdev,
 422                    sector_t sector, int size, struct page *page)
 423 {
 424         /* write first size bytes of page to sector of rdev
 425          * Increment mddev->pending_writes before returning
 426          * and decrement it on completion, waking up sb_wait
 427          * if zero is reached.
 428          * If an error occurred, call md_error
 429          *
 430          * As we might need to resubmit the request if BIO_RW_BARRIER
 431          * causes ENOTSUPP, we allocate a spare bio...
 432          */
 433         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 434         int rw = (1<<BIO_RW) | (1<<BIO_RW_SYNC);
 435
 436         bio->bi_bdev = rdev->bdev;
 437         bio->bi_sector = sector;
 438         bio_add_page(bio, page, size, 0);
 439         bio->bi_private = rdev;
 440         bio->bi_end_io = super_written;
 441         bio->bi_rw = rw;
 442
 443         atomic_inc(&mddev->pending_writes);
 444         if (!test_bit(BarriersNotsupp, &rdev->flags)) {
 445                 struct bio *rbio;
 446                 rw |= (1<<BIO_RW_BARRIER);
 447                 rbio = bio_clone(bio, GFP_NOIO);
 448                 rbio->bi_private = bio;
 449                 rbio->bi_end_io = super_written_barrier;
 450                 submit_bio(rw, rbio);
 451         } else
 452                 submit_bio(rw, bio);
 453 }
 454
 455 void md_super_wait(mddev_t *mddev)
 456 {
 457         /* wait for all superblock writes that were scheduled to complete.
 458          * if any had to be retried (due to BARRIER problems), retry them
 459          */
 460         DEFINE_WAIT(wq);
 461         for(;;) {
 462                 prepare_to_wait(&mddev->sb_wait, &wq, TASK_UNINTERRUPTIBLE);
 463                 if (atomic_read(&mddev->pending_writes)==0)
 464                         break;
 465                 while (mddev->biolist) {
 466                         struct bio *bio;
 467                         spin_lock_irq(&mddev->write_lock);
 468                         bio = mddev->biolist;
 469                         mddev->biolist = bio->bi_next ;
 470                         bio->bi_next = NULL;
 471                         spin_unlock_irq(&mddev->write_lock);
 472                         submit_bio(bio->bi_rw, bio);
 473                 }
 474                 schedule();
 475         }
 476         finish_wait(&mddev->sb_wait, &wq);
 477 }
 478
 479 static int bi_complete(struct bio *bio, unsigned int bytes_done, int error)
 480 {
 481         if (bio->bi_size)
 482                 return 1;
 483
 484         complete((struct completion*)bio->bi_private);
 485         return 0;
 486 }
 487
 488 int sync_page_io(struct block_device *bdev, sector_t sector, int size,
 489                    struct page *page, int rw)
 490 {
 491         struct bio *bio = bio_alloc(GFP_NOIO, 1);
 492         struct completion event;
 493         int ret;
 494
 495         rw |= (1 << BIO_RW_SYNC);
 496
 497         bio->bi_bdev = bdev;
 498         bio->bi_sector = sector;
 499         bio_add_page(bio, page, size, 0);
 500         init_completion(&event);
 501         bio->bi_private = &event;
 502         bio->bi_end_io = bi_complete;
 503         submit_bio(rw, bio);
 504         wait_for_completion(&event);
 505
 506         ret = test_bit(BIO_UPTODATE, &bio->bi_flags);
 507         bio_put(bio);
 508         return ret;
 509 }
 510 EXPORT_SYMBOL_GPL(sync_page_io);
 511
 512 static int read_disk_sb(mdk_rdev_t * rdev, int size)
 513 {
 514         char b[BDEVNAME_SIZE];
 515         if (!rdev->sb_page) {
 516                 MD_BUG();
 517                 return -EINVAL;
 518         }
 519         if (rdev->sb_loaded)
 520                 return 0;
 521
 522
 523         if (!sync_page_io(rdev->bdev, rdev->sb_offset<<1, size, rdev->sb_page, READ))
 524                 goto fail;
 525         rdev->sb_loaded = 1;
 526         return 0;
 527
 528 fail:
 529         printk(KERN_WARNING "md: disabled device %s, could not read superblock.\n",
 530                 bdevname(rdev->bdev,b));
 531         return -EINVAL;
 532 }
 533
 534 static int uuid_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 535 {
 536         if (    (sb1->set_uuid0 == sb2->set_uuid0) &&
 537                 (sb1->set_uuid1 == sb2->set_uuid1) &&
 538                 (sb1->set_uuid2 == sb2->set_uuid2) &&
 539                 (sb1->set_uuid3 == sb2->set_uuid3))
 540
 541                 return 1;
 542
 543         return 0;
 544 }
 545
 546
 547 static int sb_equal(mdp_super_t *sb1, mdp_super_t *sb2)
 548 {
 549         int ret;
 550         mdp_super_t *tmp1, *tmp2;
 551
 552         tmp1 = kmalloc(sizeof(*tmp1),GFP_KERNEL);
 553         tmp2 = kmalloc(sizeof(*tmp2),GFP_KERNEL);
 554
 555         if (!tmp1 || !tmp2) {
 556                 ret = 0;
 557                 printk(KERN_INFO "md.c: sb1 is not equal to sb2!\n");
 558                 goto abort;
 559         }
 560
 561         *tmp1 = *sb1;
 562         *tmp2 = *sb2;
 563
 564         /*
 565          * nr_disks is not constant
 566          */
 567         tmp1->nr_disks = 0;
 568         tmp2->nr_disks = 0;
 569
 570         if (memcmp(tmp1, tmp2, MD_SB_GENERIC_CONSTANT_WORDS * 4))
 571                 ret = 0;
 572         else
 573                 ret = 1;
 574
 575 abort:
 576         kfree(tmp1);
 577         kfree(tmp2);
 578         return ret;
 579 }
 580
 581 static unsigned int calc_sb_csum(mdp_super_t * sb)
 582 {
 583         unsigned int disk_csum, csum;
 584
 585         disk_csum = sb->sb_csum;
 586         sb->sb_csum = 0;
 587         csum = csum_partial((void *)sb, MD_SB_BYTES, 0);
 588         sb->sb_csum = disk_csum;
 589         return csum;
 590 }
 591
 592
 593 /*
 594  * Handle superblock details.
 595  * We want to be able to handle multiple superblock formats
 596  * so we have a common interface to them all, and an array of
 597  * different handlers.
 598  * We rely on user-space to write the initial superblock, and support
 599  * reading and updating of superblocks.
 600  * Interface methods are:
 601  *   int load_super(mdk_rdev_t *dev, mdk_rdev_t *refdev, int minor_version)
 602  *      loads and validates a superblock on dev.
 603  *      if refdev != NULL, compare superblocks on both devices
 604  *    Return:
 605  *      0 - dev has a superblock that is compatible with refdev
 606  *      1 - dev has a superblock that is compatible and newer than refdev
 607  *          so dev should be used as the refdev in future
 608  *     -EINVAL superblock incompatible or invalid
 609  *     -othererror e.g. -EIO
 610  *
 611  *   int validate_super(mddev_t *mddev, mdk_rdev_t *dev)
 612  *      Verify that dev is acceptable into mddev.
 613  *       The first time, mddev->raid_disks will be 0, and data from
 614  *       dev should be merged in.  Subsequent calls check that dev
 615  *       is new enough.  Return 0 or -EINVAL
 616  *
 617  *   void sync_super(mddev_t *mddev, mdk_rdev_t *dev)
 618  *     Update the superblock for rdev with data in mddev
 619  *     This does not write to disc.
 620  *
 621  */
 622
 623 struct super_type  {
 624         char            *name;
 625         struct module   *owner;
 626         int             (*load_super)(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version);
 627         int             (*validate_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 628         void            (*sync_super)(mddev_t *mddev, mdk_rdev_t *rdev);
 629 };
 630
 631 /*
 632  * load_super for 0.90.0
 633  */
 634 static int super_90_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 635 {
 636         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 637         mdp_super_t *sb;
 638         int ret;
 639         sector_t sb_offset;
 640
 641         /*
 642          * Calculate the position of the superblock,
 643          * it's at the end of the disk.
 644          *
 645          * It also happens to be a multiple of 4Kb.
 646          */
 647         sb_offset = calc_dev_sboffset(rdev->bdev);
 648         rdev->sb_offset = sb_offset;
 649
 650         ret = read_disk_sb(rdev, MD_SB_BYTES);
 651         if (ret) return ret;
 652
 653         ret = -EINVAL;
 654
 655         bdevname(rdev->bdev, b);
 656         sb = (mdp_super_t*)page_address(rdev->sb_page);
 657
 658         if (sb->md_magic != MD_SB_MAGIC) {
 659                 printk(KERN_ERR "md: invalid raid superblock magic on %s\n",
 660                        b);
 661                 goto abort;
 662         }
 663
 664         if (sb->major_version != 0 ||
 665             sb->minor_version < 90 ||
 666             sb->minor_version > 91) {
 667                 printk(KERN_WARNING "Bad version number %d.%d on %s\n",
 668                         sb->major_version, sb->minor_version,
 669                         b);
 670                 goto abort;
 671         }
 672
 673         if (sb->raid_disks <= 0)
 674                 goto abort;
 675
 676         if (csum_fold(calc_sb_csum(sb)) != csum_fold(sb->sb_csum)) {
 677                 printk(KERN_WARNING "md: invalid superblock checksum on %s\n",
 678                         b);
 679                 goto abort;
 680         }
 681
 682         rdev->preferred_minor = sb->md_minor;
 683         rdev->data_offset = 0;
 684         rdev->sb_size = MD_SB_BYTES;
 685
 686         if (sb->level == LEVEL_MULTIPATH)
 687                 rdev->desc_nr = -1;
 688         else
 689                 rdev->desc_nr = sb->this_disk.number;
 690
 691         if (refdev == 0)
 692                 ret = 1;
 693         else {
 694                 __u64 ev1, ev2;
 695                 mdp_super_t *refsb = (mdp_super_t*)page_address(refdev->sb_page);
 696                 if (!uuid_equal(refsb, sb)) {
 697                         printk(KERN_WARNING "md: %s has different UUID to %s\n",
 698                                 b, bdevname(refdev->bdev,b2));
 699                         goto abort;
 700                 }
 701                 if (!sb_equal(refsb, sb)) {
 702                         printk(KERN_WARNING "md: %s has same UUID"
 703                                " but different superblock to %s\n",
 704                                b, bdevname(refdev->bdev, b2));
 705                         goto abort;
 706                 }
 707                 ev1 = md_event(sb);
 708                 ev2 = md_event(refsb);
 709                 if (ev1 > ev2)
 710                         ret = 1;
 711                 else
 712                         ret = 0;
 713         }
 714         rdev->size = calc_dev_size(rdev, sb->chunk_size);
 715
 716         if (rdev->size < sb->size && sb->level > 1)
 717                 /* "this cannot possibly happen" ... */
 718                 ret = -EINVAL;
 719
 720  abort:
 721         return ret;
 722 }
 723
 724 /*
 725  * validate_super for 0.90.0
 726  */
 727 static int super_90_validate(mddev_t *mddev, mdk_rdev_t *rdev)
 728 {
 729         mdp_disk_t *desc;
 730         mdp_super_t *sb = (mdp_super_t *)page_address(rdev->sb_page);
 731
 732         rdev->raid_disk = -1;
 733         rdev->flags = 0;
 734         if (mddev->raid_disks == 0) {
 735                 mddev->major_version = 0;
 736                 mddev->minor_version = sb->minor_version;
 737                 mddev->patch_version = sb->patch_version;
 738                 mddev->persistent = ! sb->not_persistent;
 739                 mddev->chunk_size = sb->chunk_size;
 740                 mddev->ctime = sb->ctime;
 741                 mddev->utime = sb->utime;
 742                 mddev->level = sb->level;
 743                 mddev->clevel[0] = 0;
 744                 mddev->layout = sb->layout;
 745                 mddev->raid_disks = sb->raid_disks;
 746                 mddev->size = sb->size;
 747                 mddev->events = md_event(sb);
 748                 mddev->bitmap_offset = 0;
 749                 mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
 750
 751                 if (mddev->minor_version >= 91) {
 752                         mddev->reshape_position = sb->reshape_position;
 753                         mddev->delta_disks = sb->delta_disks;
 754                         mddev->new_level = sb->new_level;
 755                         mddev->new_layout = sb->new_layout;
 756                         mddev->new_chunk = sb->new_chunk;
 757                 } else {
 758                         mddev->reshape_position = MaxSector;
 759                         mddev->delta_disks = 0;
 760                         mddev->new_level = mddev->level;
 761                         mddev->new_layout = mddev->layout;
 762                         mddev->new_chunk = mddev->chunk_size;
 763                 }
 764
 765                 if (sb->state & (1<<MD_SB_CLEAN))
 766                         mddev->recovery_cp = MaxSector;
 767                 else {
 768                         if (sb->events_hi == sb->cp_events_hi &&
 769                                 sb->events_lo == sb->cp_events_lo) {
 770                                 mddev->recovery_cp = sb->recovery_cp;
 771                         } else
 772                                 mddev->recovery_cp = 0;
 773                 }
 774
 775                 memcpy(mddev->uuid+0, &sb->set_uuid0, 4);
 776                 memcpy(mddev->uuid+4, &sb->set_uuid1, 4);
 777                 memcpy(mddev->uuid+8, &sb->set_uuid2, 4);
 778                 memcpy(mddev->uuid+12,&sb->set_uuid3, 4);
 779
 780                 mddev->max_disks = MD_SB_DISKS;
 781
 782                 if (sb->state & (1<<MD_SB_BITMAP_PRESENT) &&
 783                     mddev->bitmap_file == NULL) {
 784                         if (mddev->level != 1 && mddev->level != 4
 785                             && mddev->level != 5 && mddev->level != 6
 786                             && mddev->level != 10) {
 787                                 /* FIXME use a better test */
 788                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
 789                                 return -EINVAL;
 790                         }
 791                         mddev->bitmap_offset = mddev->default_bitmap_offset;
 792                 }
 793
 794         } else if (mddev->pers == NULL) {
 795                 /* Insist on good event counter while assembling */
 796                 __u64 ev1 = md_event(sb);
 797                 ++ev1;
 798                 if (ev1 < mddev->events)
 799                         return -EINVAL;
 800         } else if (mddev->bitmap) {
 801                 /* if adding to array with a bitmap, then we can accept an
 802                  * older device ... but not too old.
 803                  */
 804                 __u64 ev1 = md_event(sb);
 805                 if (ev1 < mddev->bitmap->events_cleared)
 806                         return 0;
 807         } else /* just a hot-add of a new device, leave raid_disk at -1 */
 808                 return 0;
 809
 810         if (mddev->level != LEVEL_MULTIPATH) {
 811                 desc = sb->disks + rdev->desc_nr;
 812
 813                 if (desc->state & (1<<MD_DISK_FAULTY))
 814                         set_bit(Faulty, &rdev->flags);
 815                 else if (desc->state & (1<<MD_DISK_SYNC) &&
 816                          desc->raid_disk < mddev->raid_disks) {
 817                         set_bit(In_sync, &rdev->flags);
 818                         rdev->raid_disk = desc->raid_disk;
 819                 }
 820                 if (desc->state & (1<<MD_DISK_WRITEMOSTLY))
 821                         set_bit(WriteMostly, &rdev->flags);
 822         } else /* MULTIPATH are always insync */
 823                 set_bit(In_sync, &rdev->flags);
 824         return 0;
 825 }
 826
 827 /*
 828  * sync_super for 0.90.0
 829  */
 830 static void super_90_sync(mddev_t *mddev, mdk_rdev_t *rdev)
 831 {
 832         mdp_super_t *sb;
 833         struct list_head *tmp;
 834         mdk_rdev_t *rdev2;
 835         int next_spare = mddev->raid_disks;
 836
 837
 838         /* make rdev->sb match mddev data..
 839          *
 840          * 1/ zero out disks
 841          * 2/ Add info for each disk, keeping track of highest desc_nr (next_spare);
 842          * 3/ any empty disks < next_spare become removed
 843          *
 844          * disks[0] gets initialised to REMOVED because
 845          * we cannot be sure from other fields if it has
 846          * been initialised or not.
 847          */
 848         int i;
 849         int active=0, working=0,failed=0,spare=0,nr_disks=0;
 850
 851         rdev->sb_size = MD_SB_BYTES;
 852
 853         sb = (mdp_super_t*)page_address(rdev->sb_page);
 854
 855         memset(sb, 0, sizeof(*sb));
 856
 857         sb->md_magic = MD_SB_MAGIC;
 858         sb->major_version = mddev->major_version;
 859         sb->patch_version = mddev->patch_version;
 860         sb->gvalid_words  = 0; /* ignored */
 861         memcpy(&sb->set_uuid0, mddev->uuid+0, 4);
 862         memcpy(&sb->set_uuid1, mddev->uuid+4, 4);
 863         memcpy(&sb->set_uuid2, mddev->uuid+8, 4);
 864         memcpy(&sb->set_uuid3, mddev->uuid+12,4);
 865
 866         sb->ctime = mddev->ctime;
 867         sb->level = mddev->level;
 868         sb->size  = mddev->size;
 869         sb->raid_disks = mddev->raid_disks;
 870         sb->md_minor = mddev->md_minor;
 871         sb->not_persistent = !mddev->persistent;
 872         sb->utime = mddev->utime;
 873         sb->state = 0;
 874         sb->events_hi = (mddev->events>>32);
 875         sb->events_lo = (u32)mddev->events;
 876
 877         if (mddev->reshape_position == MaxSector)
 878                 sb->minor_version = 90;
 879         else {
 880                 sb->minor_version = 91;
 881                 sb->reshape_position = mddev->reshape_position;
 882                 sb->new_level = mddev->new_level;
 883                 sb->delta_disks = mddev->delta_disks;
 884                 sb->new_layout = mddev->new_layout;
 885                 sb->new_chunk = mddev->new_chunk;
 886         }
 887         mddev->minor_version = sb->minor_version;
 888         if (mddev->in_sync)
 889         {
 890                 sb->recovery_cp = mddev->recovery_cp;
 891                 sb->cp_events_hi = (mddev->events>>32);
 892                 sb->cp_events_lo = (u32)mddev->events;
 893                 if (mddev->recovery_cp == MaxSector)
 894                         sb->state = (1<< MD_SB_CLEAN);
 895         } else
 896                 sb->recovery_cp = 0;
 897
 898         sb->layout = mddev->layout;
 899         sb->chunk_size = mddev->chunk_size;
 900
 901         if (mddev->bitmap && mddev->bitmap_file == NULL)
 902                 sb->state |= (1<<MD_SB_BITMAP_PRESENT);
 903
 904         sb->disks[0].state = (1<<MD_DISK_REMOVED);
 905         ITERATE_RDEV(mddev,rdev2,tmp) {
 906                 mdp_disk_t *d;
 907                 int desc_nr;
 908                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 909                     && !test_bit(Faulty, &rdev2->flags))
 910                         desc_nr = rdev2->raid_disk;
 911                 else
 912                         desc_nr = next_spare++;
 913                 rdev2->desc_nr = desc_nr;
 914                 d = &sb->disks[rdev2->desc_nr];
 915                 nr_disks++;
 916                 d->number = rdev2->desc_nr;
 917                 d->major = MAJOR(rdev2->bdev->bd_dev);
 918                 d->minor = MINOR(rdev2->bdev->bd_dev);
 919                 if (rdev2->raid_disk >= 0 && test_bit(In_sync, &rdev2->flags)
 920                     && !test_bit(Faulty, &rdev2->flags))
 921                         d->raid_disk = rdev2->raid_disk;
 922                 else
 923                         d->raid_disk = rdev2->desc_nr; /* compatibility */
 924                 if (test_bit(Faulty, &rdev2->flags))
 925                         d->state = (1<<MD_DISK_FAULTY);
 926                 else if (test_bit(In_sync, &rdev2->flags)) {
 927                         d->state = (1<<MD_DISK_ACTIVE);
 928                         d->state |= (1<<MD_DISK_SYNC);
 929                         active++;
 930                         working++;
 931                 } else {
 932                         d->state = 0;
 933                         spare++;
 934                         working++;
 935                 }
 936                 if (test_bit(WriteMostly, &rdev2->flags))
 937                         d->state |= (1<<MD_DISK_WRITEMOSTLY);
 938         }
 939         /* now set the "removed" and "faulty" bits on any missing devices */
 940         for (i=0 ; i < mddev->raid_disks ; i++) {
 941                 mdp_disk_t *d = &sb->disks[i];
 942                 if (d->state == 0 && d->number == 0) {
 943                         d->number = i;
 944                         d->raid_disk = i;
 945                         d->state = (1<<MD_DISK_REMOVED);
 946                         d->state |= (1<<MD_DISK_FAULTY);
 947                         failed++;
 948                 }
 949         }
 950         sb->nr_disks = nr_disks;
 951         sb->active_disks = active;
 952         sb->working_disks = working;
 953         sb->failed_disks = failed;
 954         sb->spare_disks = spare;
 955
 956         sb->this_disk = sb->disks[rdev->desc_nr];
 957         sb->sb_csum = calc_sb_csum(sb);
 958 }
 959
 960 /*
 961  * version 1 superblock
 962  */
 963
 964 static unsigned int calc_sb_1_csum(struct mdp_superblock_1 * sb)
 965 {
 966         unsigned int disk_csum, csum;
 967         unsigned long long newcsum;
 968         int size = 256 + le32_to_cpu(sb->max_dev)*2;
 969         unsigned int *isuper = (unsigned int*)sb;
 970         int i;
 971
 972         disk_csum = sb->sb_csum;
 973         sb->sb_csum = 0;
 974         newcsum = 0;
 975         for (i=0; size>=4; size -= 4 )
 976                 newcsum += le32_to_cpu(*isuper++);
 977
 978         if (size == 2)
 979                 newcsum += le16_to_cpu(*(unsigned short*) isuper);
 980
 981         csum = (newcsum & 0xffffffff) + (newcsum >> 32);
 982         sb->sb_csum = disk_csum;
 983         return cpu_to_le32(csum);
 984 }
 985
 986 static int super_1_load(mdk_rdev_t *rdev, mdk_rdev_t *refdev, int minor_version)
 987 {
 988         struct mdp_superblock_1 *sb;
 989         int ret;
 990         sector_t sb_offset;
 991         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
 992         int bmask;
 993
 994         /*
 995          * Calculate the position of the superblock.
 996          * It is always aligned to a 4K boundary and
 997          * depeding on minor_version, it can be:
 998          * 0: At least 8K, but less than 12K, from end of device
 999          * 1: At start of device
1000          * 2: 4K from start of device.
1001          */
1002         switch(minor_version) {
1003         case 0:
1004                 sb_offset = rdev->bdev->bd_inode->i_size >> 9;
1005                 sb_offset -= 8*2;
1006                 sb_offset &= ~(sector_t)(4*2-1);
1007                 /* convert from sectors to K */
1008                 sb_offset /= 2;
1009                 break;
1010         case 1:
1011                 sb_offset = 0;
1012                 break;
1013         case 2:
1014                 sb_offset = 4;
1015                 break;
1016         default:
1017                 return -EINVAL;
1018         }
1019         rdev->sb_offset = sb_offset;
1020
1021         /* superblock is rarely larger than 1K, but it can be larger,
1022          * and it is safe to read 4k, so we do that
1023          */
1024         ret = read_disk_sb(rdev, 4096);
1025         if (ret) return ret;
1026
1027
1028         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1029
1030         if (sb->magic != cpu_to_le32(MD_SB_MAGIC) ||
1031             sb->major_version != cpu_to_le32(1) ||
1032             le32_to_cpu(sb->max_dev) > (4096-256)/2 ||
1033             le64_to_cpu(sb->super_offset) != (rdev->sb_offset<<1) ||
1034             (le32_to_cpu(sb->feature_map) & ~MD_FEATURE_ALL) != 0)
1035                 return -EINVAL;
1036
1037         if (calc_sb_1_csum(sb) != sb->sb_csum) {
1038                 printk("md: invalid superblock checksum on %s\n",
1039                         bdevname(rdev->bdev,b));
1040                 return -EINVAL;
1041         }
1042         if (le64_to_cpu(sb->data_size) < 10) {
1043                 printk("md: data_size too small on %s\n",
1044                        bdevname(rdev->bdev,b));
1045                 return -EINVAL;
1046         }
1047         rdev->preferred_minor = 0xffff;
1048         rdev->data_offset = le64_to_cpu(sb->data_offset);
1049         atomic_set(&rdev->corrected_errors, le32_to_cpu(sb->cnt_corrected_read));
1050
1051         rdev->sb_size = le32_to_cpu(sb->max_dev) * 2 + 256;
1052         bmask = queue_hardsect_size(rdev->bdev->bd_disk->queue)-1;
1053         if (rdev->sb_size & bmask)
1054                 rdev-> sb_size = (rdev->sb_size | bmask)+1;
1055
1056         if (refdev == 0)
1057                 ret = 1;
1058         else {
1059                 __u64 ev1, ev2;
1060                 struct mdp_superblock_1 *refsb =
1061                         (struct mdp_superblock_1*)page_address(refdev->sb_page);
1062
1063                 if (memcmp(sb->set_uuid, refsb->set_uuid, 16) != 0 ||
1064                     sb->level != refsb->level ||
1065                     sb->layout != refsb->layout ||
1066                     sb->chunksize != refsb->chunksize) {
1067                         printk(KERN_WARNING "md: %s has strangely different"
1068                                 " superblock to %s\n",
1069                                 bdevname(rdev->bdev,b),
1070                                 bdevname(refdev->bdev,b2));
1071                         return -EINVAL;
1072                 }
1073                 ev1 = le64_to_cpu(sb->events);
1074                 ev2 = le64_to_cpu(refsb->events);
1075
1076                 if (ev1 > ev2)
1077                         ret = 1;
1078                 else
1079                         ret = 0;
1080         }
1081         if (minor_version)
1082                 rdev->size = ((rdev->bdev->bd_inode->i_size>>9) - le64_to_cpu(sb->data_offset)) / 2;
1083         else
1084                 rdev->size = rdev->sb_offset;
1085         if (rdev->size < le64_to_cpu(sb->data_size)/2)
1086                 return -EINVAL;
1087         rdev->size = le64_to_cpu(sb->data_size)/2;
1088         if (le32_to_cpu(sb->chunksize))
1089                 rdev->size &= ~((sector_t)le32_to_cpu(sb->chunksize)/2 - 1);
1090
1091         if (le32_to_cpu(sb->size) > rdev->size*2)
1092                 return -EINVAL;
1093         return ret;
1094 }
1095
1096 static int super_1_validate(mddev_t *mddev, mdk_rdev_t *rdev)
1097 {
1098         struct mdp_superblock_1 *sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1099
1100         rdev->raid_disk = -1;
1101         rdev->flags = 0;
1102         if (mddev->raid_disks == 0) {
1103                 mddev->major_version = 1;
1104                 mddev->patch_version = 0;
1105                 mddev->persistent = 1;
1106                 mddev->chunk_size = le32_to_cpu(sb->chunksize) << 9;
1107                 mddev->ctime = le64_to_cpu(sb->ctime) & ((1ULL << 32)-1);
1108                 mddev->utime = le64_to_cpu(sb->utime) & ((1ULL << 32)-1);
1109                 mddev->level = le32_to_cpu(sb->level);
1110                 mddev->clevel[0] = 0;
1111                 mddev->layout = le32_to_cpu(sb->layout);
1112                 mddev->raid_disks = le32_to_cpu(sb->raid_disks);
1113                 mddev->size = le64_to_cpu(sb->size)/2;
1114                 mddev->events = le64_to_cpu(sb->events);
1115                 mddev->bitmap_offset = 0;
1116                 mddev->default_bitmap_offset = 1024 >> 9;
1117
1118                 mddev->recovery_cp = le64_to_cpu(sb->resync_offset);
1119                 memcpy(mddev->uuid, sb->set_uuid, 16);
1120
1121                 mddev->max_disks =  (4096-256)/2;
1122
1123                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_BITMAP_OFFSET) &&
1124                     mddev->bitmap_file == NULL ) {
1125                         if (mddev->level != 1 && mddev->level != 5 && mddev->level != 6
1126                             && mddev->level != 10) {
1127                                 printk(KERN_WARNING "md: bitmaps not supported for this level.\n");
1128                                 return -EINVAL;
1129                         }
1130                         mddev->bitmap_offset = (__s32)le32_to_cpu(sb->bitmap_offset);
1131                 }
1132                 if ((le32_to_cpu(sb->feature_map) & MD_FEATURE_RESHAPE_ACTIVE)) {
1133                         mddev->reshape_position = le64_to_cpu(sb->reshape_position);
1134                         mddev->delta_disks = le32_to_cpu(sb->delta_disks);
1135                         mddev->new_level = le32_to_cpu(sb->new_level);
1136                         mddev->new_layout = le32_to_cpu(sb->new_layout);
1137                         mddev->new_chunk = le32_to_cpu(sb->new_chunk)<<9;
1138                 } else {
1139                         mddev->reshape_position = MaxSector;
1140                         mddev->delta_disks = 0;
1141                         mddev->new_level = mddev->level;
1142                         mddev->new_layout = mddev->layout;
1143                         mddev->new_chunk = mddev->chunk_size;
1144                 }
1145
1146         } else if (mddev->pers == NULL) {
1147                 /* Insist of good event counter while assembling */
1148                 __u64 ev1 = le64_to_cpu(sb->events);
1149                 ++ev1;
1150                 if (ev1 < mddev->events)
1151                         return -EINVAL;
1152         } else if (mddev->bitmap) {
1153                 /* If adding to array with a bitmap, then we can accept an
1154                  * older device, but not too old.
1155                  */
1156                 __u64 ev1 = le64_to_cpu(sb->events);
1157                 if (ev1 < mddev->bitmap->events_cleared)
1158                         return 0;
1159         } else /* just a hot-add of a new device, leave raid_disk at -1 */
1160                 return 0;
1161
1162         if (mddev->level != LEVEL_MULTIPATH) {
1163                 int role;
1164                 rdev->desc_nr = le32_to_cpu(sb->dev_number);
1165                 role = le16_to_cpu(sb->dev_roles[rdev->desc_nr]);
1166                 switch(role) {
1167                 case 0xffff: /* spare */
1168                         break;
1169                 case 0xfffe: /* faulty */
1170                         set_bit(Faulty, &rdev->flags);
1171                         break;
1172                 default:
1173                         set_bit(In_sync, &rdev->flags);
1174                         rdev->raid_disk = role;
1175                         break;
1176                 }
1177                 if (sb->devflags & WriteMostly1)
1178                         set_bit(WriteMostly, &rdev->flags);
1179         } else /* MULTIPATH are always insync */
1180                 set_bit(In_sync, &rdev->flags);
1181
1182         return 0;
1183 }
1184
1185 static void super_1_sync(mddev_t *mddev, mdk_rdev_t *rdev)
1186 {
1187         struct mdp_superblock_1 *sb;
1188         struct list_head *tmp;
1189         mdk_rdev_t *rdev2;
1190         int max_dev, i;
1191         /* make rdev->sb match mddev and rdev data. */
1192
1193         sb = (struct mdp_superblock_1*)page_address(rdev->sb_page);
1194
1195         sb->feature_map = 0;
1196         sb->pad0 = 0;
1197         memset(sb->pad1, 0, sizeof(sb->pad1));
1198         memset(sb->pad2, 0, sizeof(sb->pad2));
1199         memset(sb->pad3, 0, sizeof(sb->pad3));
1200
1201         sb->utime = cpu_to_le64((__u64)mddev->utime);
1202         sb->events = cpu_to_le64(mddev->events);
1203         if (mddev->in_sync)
1204                 sb->resync_offset = cpu_to_le64(mddev->recovery_cp);
1205         else
1206                 sb->resync_offset = cpu_to_le64(0);
1207
1208         sb->cnt_corrected_read = atomic_read(&rdev->corrected_errors);
1209
1210         sb->raid_disks = cpu_to_le32(mddev->raid_disks);
1211         sb->size = cpu_to_le64(mddev->size<<1);
1212
1213         if (mddev->bitmap && mddev->bitmap_file == NULL) {
1214                 sb->bitmap_offset = cpu_to_le32((__u32)mddev->bitmap_offset);
1215                 sb->feature_map = cpu_to_le32(MD_FEATURE_BITMAP_OFFSET);
1216         }
1217         if (mddev->reshape_position != MaxSector) {
1218                 sb->feature_map |= cpu_to_le32(MD_FEATURE_RESHAPE_ACTIVE);
1219                 sb->reshape_position = cpu_to_le64(mddev->reshape_position);
1220                 sb->new_layout = cpu_to_le32(mddev->new_layout);
1221                 sb->delta_disks = cpu_to_le32(mddev->delta_disks);
1222                 sb->new_level = cpu_to_le32(mddev->new_level);
1223                 sb->new_chunk = cpu_to_le32(mddev->new_chunk>>9);
1224         }
1225
1226         max_dev = 0;
1227         ITERATE_RDEV(mddev,rdev2,tmp)
1228                 if (rdev2->desc_nr+1 > max_dev)
1229                         max_dev = rdev2->desc_nr+1;
1230
1231         sb->max_dev = cpu_to_le32(max_dev);
1232         for (i=0; i<max_dev;i++)
1233                 sb->dev_roles[i] = cpu_to_le16(0xfffe);
1234
1235         ITERATE_RDEV(mddev,rdev2,tmp) {
1236                 i = rdev2->desc_nr;
1237                 if (test_bit(Faulty, &rdev2->flags))
1238                         sb->dev_roles[i] = cpu_to_le16(0xfffe);
1239                 else if (test_bit(In_sync, &rdev2->flags))
1240                         sb->dev_roles[i] = cpu_to_le16(rdev2->raid_disk);
1241                 else
1242                         sb->dev_roles[i] = cpu_to_le16(0xffff);
1243         }
1244
1245         sb->recovery_offset = cpu_to_le64(0); /* not supported yet */
1246         sb->sb_csum = calc_sb_1_csum(sb);
1247 }
1248
1249
1250 static struct super_type super_types[] = {
1251         [0] = {
1252                 .name   = "0.90.0",
1253                 .owner  = THIS_MODULE,
1254                 .load_super     = super_90_load,
1255                 .validate_super = super_90_validate,
1256                 .sync_super     = super_90_sync,
1257         },
1258         [1] = {
1259                 .name   = "md-1",
1260                 .owner  = THIS_MODULE,
1261                 .load_super     = super_1_load,
1262                 .validate_super = super_1_validate,
1263                 .sync_super     = super_1_sync,
1264         },
1265 };
1266
1267 static mdk_rdev_t * match_dev_unit(mddev_t *mddev, mdk_rdev_t *dev)
1268 {
1269         struct list_head *tmp;
1270         mdk_rdev_t *rdev;
1271
1272         ITERATE_RDEV(mddev,rdev,tmp)
1273                 if (rdev->bdev->bd_contains == dev->bdev->bd_contains)
1274                         return rdev;
1275
1276         return NULL;
1277 }
1278
1279 static int match_mddev_units(mddev_t *mddev1, mddev_t *mddev2)
1280 {
1281         struct list_head *tmp;
1282         mdk_rdev_t *rdev;
1283
1284         ITERATE_RDEV(mddev1,rdev,tmp)
1285                 if (match_dev_unit(mddev2, rdev))
1286                         return 1;
1287
1288         return 0;
1289 }
1290
1291 static LIST_HEAD(pending_raid_disks);
1292
1293 static int bind_rdev_to_array(mdk_rdev_t * rdev, mddev_t * mddev)
1294 {
1295         mdk_rdev_t *same_pdev;
1296         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
1297         struct kobject *ko;
1298         char *s;
1299
1300         if (rdev->mddev) {
1301                 MD_BUG();
1302                 return -EINVAL;
1303         }
1304         /* make sure rdev->size exceeds mddev->size */
1305         if (rdev->size && (mddev->size == 0 || rdev->size < mddev->size)) {
1306                 if (mddev->pers)
1307                         /* Cannot change size, so fail */
1308                         return -ENOSPC;
1309                 else
1310                         mddev->size = rdev->size;
1311         }
1312         same_pdev = match_dev_unit(mddev, rdev);
1313         if (same_pdev)
1314                 printk(KERN_WARNING
1315                         "%s: WARNING: %s appears to be on the same physical"
1316                         " disk as %s. True\n     protection against single-disk"
1317                         " failure might be compromised.\n",
1318                         mdname(mddev), bdevname(rdev->bdev,b),
1319                         bdevname(same_pdev->bdev,b2));
1320
1321         /* Verify rdev->desc_nr is unique.
1322          * If it is -1, assign a free number, else
1323          * check number is not in use
1324          */
1325         if (rdev->desc_nr < 0) {
1326                 int choice = 0;
1327                 if (mddev->pers) choice = mddev->raid_disks;
1328                 while (find_rdev_nr(mddev, choice))
1329                         choice++;
1330                 rdev->desc_nr = choice;
1331         } else {
1332                 if (find_rdev_nr(mddev, rdev->desc_nr))
1333                         return -EBUSY;
1334         }
1335         bdevname(rdev->bdev,b);
1336         if (kobject_set_name(&rdev->kobj, "dev-%s", b) < 0)
1337                 return -ENOMEM;
1338         while ( (s=strchr(rdev->kobj.k_name, '/')) != NULL)
1339                 *s = '!';
1340
1341         list_add(&rdev->same_set, &mddev->disks);
1342         rdev->mddev = mddev;
1343         printk(KERN_INFO "md: bind<%s>\n", b);
1344
1345         rdev->kobj.parent = &mddev->kobj;
1346         kobject_add(&rdev->kobj);
1347
1348         if (rdev->bdev->bd_part)
1349                 ko = &rdev->bdev->bd_part->kobj;
1350         else
1351                 ko = &rdev->bdev->bd_disk->kobj;
1352         sysfs_create_link(&rdev->kobj, ko, "block");
1353         bd_claim_by_disk(rdev->bdev, rdev, mddev->gendisk);
1354         return 0;
1355 }
1356
1357 static void unbind_rdev_from_array(mdk_rdev_t * rdev)
1358 {
1359         char b[BDEVNAME_SIZE];
1360         if (!rdev->mddev) {
1361                 MD_BUG();
1362                 return;
1363         }
1364         bd_release_from_disk(rdev->bdev, rdev->mddev->gendisk);
1365         list_del_init(&rdev->same_set);
1366         printk(KERN_INFO "md: unbind<%s>\n", bdevname(rdev->bdev,b));
1367         rdev->mddev = NULL;
1368         sysfs_remove_link(&rdev->kobj, "block");
1369         kobject_del(&rdev->kobj);
1370 }
1371
1372 /*
1373  * prevent the device from being mounted, repartitioned or
1374  * otherwise reused by a RAID array (or any other kernel
1375  * subsystem), by bd_claiming the device.
1376  */
1377 static int lock_rdev(mdk_rdev_t *rdev, dev_t dev)
1378 {
1379         int err = 0;
1380         struct block_device *bdev;
1381         char b[BDEVNAME_SIZE];
1382
1383         bdev = open_by_devnum(dev, FMODE_READ|FMODE_WRITE);
1384         if (IS_ERR(bdev)) {
1385                 printk(KERN_ERR "md: could not open %s.\n",
1386                         __bdevname(dev, b));
1387                 return PTR_ERR(bdev);
1388         }
1389         err = bd_claim(bdev, rdev);
1390         if (err) {
1391                 printk(KERN_ERR "md: could not bd_claim %s.\n",
1392                         bdevname(bdev, b));
1393                 blkdev_put(bdev);
1394                 return err;
1395         }
1396         rdev->bdev = bdev;
1397         return err;
1398 }
1399
1400 static void unlock_rdev(mdk_rdev_t *rdev)
1401 {
1402         struct block_device *bdev = rdev->bdev;
1403         rdev->bdev = NULL;
1404         if (!bdev)
1405                 MD_BUG();
1406         bd_release(bdev);
1407         blkdev_put(bdev);
1408 }
1409
1410 void md_autodetect_dev(dev_t dev);
1411
1412 static void export_rdev(mdk_rdev_t * rdev)
1413 {
1414         char b[BDEVNAME_SIZE];
1415         printk(KERN_INFO "md: export_rdev(%s)\n",
1416                 bdevname(rdev->bdev,b));
1417         if (rdev->mddev)
1418                 MD_BUG();
1419         free_disk_sb(rdev);
1420         list_del_init(&rdev->same_set);
1421 #ifndef MODULE
1422         md_autodetect_dev(rdev->bdev->bd_dev);
1423 #endif
1424         unlock_rdev(rdev);
1425         kobject_put(&rdev->kobj);
1426 }
1427
1428 static void kick_rdev_from_array(mdk_rdev_t * rdev)
1429 {
1430         unbind_rdev_from_array(rdev);
1431         export_rdev(rdev);
1432 }
1433
1434 static void export_array(mddev_t *mddev)
1435 {
1436         struct list_head *tmp;
1437         mdk_rdev_t *rdev;
1438
1439         ITERATE_RDEV(mddev,rdev,tmp) {
1440                 if (!rdev->mddev) {
1441                         MD_BUG();
1442                         continue;
1443                 }
1444                 kick_rdev_from_array(rdev);
1445         }
1446         if (!list_empty(&mddev->disks))
1447                 MD_BUG();
1448         mddev->raid_disks = 0;
1449         mddev->major_version = 0;
1450 }
1451
1452 static void print_desc(mdp_disk_t *desc)
1453 {
1454         printk(" DISK<N:%d,(%d,%d),R:%d,S:%d>\n", desc->number,
1455                 desc->major,desc->minor,desc->raid_disk,desc->state);
1456 }
1457
1458 static void print_sb(mdp_super_t *sb)
1459 {
1460         int i;
1461
1462         printk(KERN_INFO
1463                 "md:  SB: (V:%d.%d.%d) ID:<%08x.%08x.%08x.%08x> CT:%08x\n",
1464                 sb->major_version, sb->minor_version, sb->patch_version,
1465                 sb->set_uuid0, sb->set_uuid1, sb->set_uuid2, sb->set_uuid3,
1466                 sb->ctime);
1467         printk(KERN_INFO "md:     L%d S%08d ND:%d RD:%d md%d LO:%d CS:%d\n",
1468                 sb->level, sb->size, sb->nr_disks, sb->raid_disks,
1469                 sb->md_minor, sb->layout, sb->chunk_size);
1470         printk(KERN_INFO "md:     UT:%08x ST:%d AD:%d WD:%d"
1471                 " FD:%d SD:%d CSUM:%08x E:%08lx\n",
1472                 sb->utime, sb->state, sb->active_disks, sb->working_disks,
1473                 sb->failed_disks, sb->spare_disks,
1474                 sb->sb_csum, (unsigned long)sb->events_lo);
1475
1476         printk(KERN_INFO);
1477         for (i = 0; i < MD_SB_DISKS; i++) {
1478                 mdp_disk_t *desc;
1479
1480                 desc = sb->disks + i;
1481                 if (desc->number || desc->major || desc->minor ||
1482                     desc->raid_disk || (desc->state && (desc->state != 4))) {
1483                         printk("     D %2d: ", i);
1484                         print_desc(desc);
1485                 }
1486         }
1487         printk(KERN_INFO "md:     THIS: ");
1488         print_desc(&sb->this_disk);
1489
1490 }
1491
1492 static void print_rdev(mdk_rdev_t *rdev)
1493 {
1494         char b[BDEVNAME_SIZE];
1495         printk(KERN_INFO "md: rdev %s, SZ:%08llu F:%d S:%d DN:%u\n",
1496                 bdevname(rdev->bdev,b), (unsigned long long)rdev->size,
1497                 test_bit(Faulty, &rdev->flags), test_bit(In_sync, &rdev->flags),
1498                 rdev->desc_nr);
1499         if (rdev->sb_loaded) {
1500                 printk(KERN_INFO "md: rdev superblock:\n");
1501                 print_sb((mdp_super_t*)page_address(rdev->sb_page));
1502         } else
1503                 printk(KERN_INFO "md: no rdev superblock!\n");
1504 }
1505
1506 void md_print_devices(void)
1507 {
1508         struct list_head *tmp, *tmp2;
1509         mdk_rdev_t *rdev;
1510         mddev_t *mddev;
1511         char b[BDEVNAME_SIZE];
1512
1513         printk("\n");
1514         printk("md:     **********************************\n");
1515         printk("md:     * <COMPLETE RAID STATE PRINTOUT> *\n");
1516         printk("md:     **********************************\n");
1517         ITERATE_MDDEV(mddev,tmp) {
1518
1519                 if (mddev->bitmap)
1520                         bitmap_print_sb(mddev->bitmap);
1521                 else
1522                         printk("%s: ", mdname(mddev));
1523                 ITERATE_RDEV(mddev,rdev,tmp2)
1524                         printk("<%s>", bdevname(rdev->bdev,b));
1525                 printk("\n");
1526
1527                 ITERATE_RDEV(mddev,rdev,tmp2)
1528                         print_rdev(rdev);
1529         }
1530         printk("md:     **********************************\n");
1531         printk("\n");
1532 }
1533
1534
1535 static void sync_sbs(mddev_t * mddev)
1536 {
1537         mdk_rdev_t *rdev;
1538         struct list_head *tmp;
1539
1540         ITERATE_RDEV(mddev,rdev,tmp) {
1541                 super_types[mddev->major_version].
1542                         sync_super(mddev, rdev);
1543                 rdev->sb_loaded = 1;
1544         }
1545 }
1546
1547 void md_update_sb(mddev_t * mddev)
1548 {
1549         int err;
1550         struct list_head *tmp;
1551         mdk_rdev_t *rdev;
1552         int sync_req;
1553
1554 repeat:
1555         spin_lock_irq(&mddev->write_lock);
1556         sync_req = mddev->in_sync;
1557         mddev->utime = get_seconds();
1558         mddev->events ++;
1559
1560         if (!mddev->events) {
1561                 /*
1562                  * oops, this 64-bit counter should never wrap.
1563                  * Either we are in around ~1 trillion A.C., assuming
1564                  * 1 reboot per second, or we have a bug:
1565                  */
1566                 MD_BUG();
1567                 mddev->events --;
1568         }
1569         mddev->sb_dirty = 2;
1570         sync_sbs(mddev);
1571
1572         /*
1573          * do not write anything to disk if using
1574          * nonpersistent superblocks
1575          */
1576         if (!mddev->persistent) {
1577                 mddev->sb_dirty = 0;
1578                 spin_unlock_irq(&mddev->write_lock);
1579                 wake_up(&mddev->sb_wait);
1580                 return;
1581         }
1582         spin_unlock_irq(&mddev->write_lock);
1583
1584         dprintk(KERN_INFO
1585                 "md: updating %s RAID superblock on device (in sync %d)\n",
1586                 mdname(mddev),mddev->in_sync);
1587
1588         err = bitmap_update_sb(mddev->bitmap);
1589         ITERATE_RDEV(mddev,rdev,tmp) {
1590                 char b[BDEVNAME_SIZE];
1591                 dprintk(KERN_INFO "md: ");
1592                 if (test_bit(Faulty, &rdev->flags))
1593                         dprintk("(skipping faulty ");
1594
1595                 dprintk("%s ", bdevname(rdev->bdev,b));
1596                 if (!test_bit(Faulty, &rdev->flags)) {
1597                         md_super_write(mddev,rdev,
1598                                        rdev->sb_offset<<1, rdev->sb_size,
1599                                        rdev->sb_page);
1600                         dprintk(KERN_INFO "(write) %s's sb offset: %llu\n",
1601                                 bdevname(rdev->bdev,b),
1602                                 (unsigned long long)rdev->sb_offset);
1603
1604                 } else
1605                         dprintk(")\n");
1606                 if (mddev->level == LEVEL_MULTIPATH)
1607                         /* only need to write one superblock... */
1608                         break;
1609         }
1610         md_super_wait(mddev);
1611         /* if there was a failure, sb_dirty was set to 1, and we re-write super */
1612
1613         spin_lock_irq(&mddev->write_lock);
1614         if (mddev->in_sync != sync_req|| mddev->sb_dirty == 1) {
1615                 /* have to write it out again */
1616                 spin_unlock_irq(&mddev->write_lock);
1617                 goto repeat;
1618         }
1619         mddev->sb_dirty = 0;
1620         spin_unlock_irq(&mddev->write_lock);
1621         wake_up(&mddev->sb_wait);
1622
1623 }
1624 EXPORT_SYMBOL_GPL(md_update_sb);
1625
1626 /* words written to sysfs files may, or my not, be \n terminated.
1627  * We want to accept with case. For this we use cmd_match.
1628  */
1629 static int cmd_match(const char *cmd, const char *str)
1630 {
1631         /* See if cmd, written into a sysfs file, matches
1632          * str.  They must either be the same, or cmd can
1633          * have a trailing newline
1634          */
1635         while (*cmd && *str && *cmd == *str) {
1636                 cmd++;
1637                 str++;
1638         }
1639         if (*cmd == '\n')
1640                 cmd++;
1641         if (*str || *cmd)
1642                 return 0;
1643         return 1;
1644 }
1645
1646 struct rdev_sysfs_entry {
1647         struct attribute attr;
1648         ssize_t (*show)(mdk_rdev_t *, char *);
1649         ssize_t (*store)(mdk_rdev_t *, const char *, size_t);
1650 };
1651
1652 static ssize_t
1653 state_show(mdk_rdev_t *rdev, char *page)
1654 {
1655         char *sep = "";
1656         int len=0;
1657
1658         if (test_bit(Faulty, &rdev->flags)) {
1659                 len+= sprintf(page+len, "%sfaulty",sep);
1660                 sep = ",";
1661         }
1662         if (test_bit(In_sync, &rdev->flags)) {
1663                 len += sprintf(page+len, "%sin_sync",sep);
1664                 sep = ",";
1665         }
1666         if (!test_bit(Faulty, &rdev->flags) &&
1667             !test_bit(In_sync, &rdev->flags)) {
1668                 len += sprintf(page+len, "%sspare", sep);
1669                 sep = ",";
1670         }
1671         return len+sprintf(page+len, "\n");
1672 }
1673
1674 static struct rdev_sysfs_entry
1675 rdev_state = __ATTR_RO(state);
1676
1677 static ssize_t
1678 super_show(mdk_rdev_t *rdev, char *page)
1679 {
1680         if (rdev->sb_loaded && rdev->sb_size) {
1681                 memcpy(page, page_address(rdev->sb_page), rdev->sb_size);
1682                 return rdev->sb_size;
1683         } else
1684                 return 0;
1685 }
1686 static struct rdev_sysfs_entry rdev_super = __ATTR_RO(super);
1687
1688 static ssize_t
1689 errors_show(mdk_rdev_t *rdev, char *page)
1690 {
1691         return sprintf(page, "%d\n", atomic_read(&rdev->corrected_errors));
1692 }
1693
1694 static ssize_t
1695 errors_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1696 {
1697         char *e;
1698         unsigned long n = simple_strtoul(buf, &e, 10);
1699         if (*buf && (*e == 0 || *e == '\n')) {
1700                 atomic_set(&rdev->corrected_errors, n);
1701                 return len;
1702         }
1703         return -EINVAL;
1704 }
1705 static struct rdev_sysfs_entry rdev_errors =
1706 __ATTR(errors, 0644, errors_show, errors_store);
1707
1708 static ssize_t
1709 slot_show(mdk_rdev_t *rdev, char *page)
1710 {
1711         if (rdev->raid_disk < 0)
1712                 return sprintf(page, "none\n");
1713         else
1714                 return sprintf(page, "%d\n", rdev->raid_disk);
1715 }
1716
1717 static ssize_t
1718 slot_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1719 {
1720         char *e;
1721         int slot = simple_strtoul(buf, &e, 10);
1722         if (strncmp(buf, "none", 4)==0)
1723                 slot = -1;
1724         else if (e==buf || (*e && *e!= '\n'))
1725                 return -EINVAL;
1726         if (rdev->mddev->pers)
1727                 /* Cannot set slot in active array (yet) */
1728                 return -EBUSY;
1729         if (slot >= rdev->mddev->raid_disks)
1730                 return -ENOSPC;
1731         rdev->raid_disk = slot;
1732         /* assume it is working */
1733         rdev->flags = 0;
1734         set_bit(In_sync, &rdev->flags);
1735         return len;
1736 }
1737
1738
1739 static struct rdev_sysfs_entry rdev_slot =
1740 __ATTR(slot, 0644, slot_show, slot_store);
1741
1742 static ssize_t
1743 offset_show(mdk_rdev_t *rdev, char *page)
1744 {
1745         return sprintf(page, "%llu\n", (unsigned long long)rdev->data_offset);
1746 }
1747
1748 static ssize_t
1749 offset_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1750 {
1751         char *e;
1752         unsigned long long offset = simple_strtoull(buf, &e, 10);
1753         if (e==buf || (*e && *e != '\n'))
1754                 return -EINVAL;
1755         if (rdev->mddev->pers)
1756                 return -EBUSY;
1757         rdev->data_offset = offset;
1758         return len;
1759 }
1760
1761 static struct rdev_sysfs_entry rdev_offset =
1762 __ATTR(offset, 0644, offset_show, offset_store);
1763
1764 static ssize_t
1765 rdev_size_show(mdk_rdev_t *rdev, char *page)
1766 {
1767         return sprintf(page, "%llu\n", (unsigned long long)rdev->size);
1768 }
1769
1770 static ssize_t
1771 rdev_size_store(mdk_rdev_t *rdev, const char *buf, size_t len)
1772 {
1773         char *e;
1774         unsigned long long size = simple_strtoull(buf, &e, 10);
1775         if (e==buf || (*e && *e != '\n'))
1776                 return -EINVAL;
1777         if (rdev->mddev->pers)
1778                 return -EBUSY;
1779         rdev->size = size;
1780         if (size < rdev->mddev->size || rdev->mddev->size == 0)
1781                 rdev->mddev->size = size;
1782         return len;
1783 }
1784
1785 static struct rdev_sysfs_entry rdev_size =
1786 __ATTR(size, 0644, rdev_size_show, rdev_size_store);
1787
1788 static struct attribute *rdev_default_attrs[] = {
1789         &rdev_state.attr,
1790         &rdev_super.attr,
1791         &rdev_errors.attr,
1792         &rdev_slot.attr,
1793         &rdev_offset.attr,
1794         &rdev_size.attr,
1795         NULL,
1796 };
1797 static ssize_t
1798 rdev_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
1799 {
1800         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1801         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1802
1803         if (!entry->show)
1804                 return -EIO;
1805         return entry->show(rdev, page);
1806 }
1807
1808 static ssize_t
1809 rdev_attr_store(struct kobject *kobj, struct attribute *attr,
1810               const char *page, size_t length)
1811 {
1812         struct rdev_sysfs_entry *entry = container_of(attr, struct rdev_sysfs_entry, attr);
1813         mdk_rdev_t *rdev = container_of(kobj, mdk_rdev_t, kobj);
1814
1815         if (!entry->store)
1816                 return -EIO;
1817         return entry->store(rdev, page, length);
1818 }
1819
1820 static void rdev_free(struct kobject *ko)
1821 {
1822         mdk_rdev_t *rdev = container_of(ko, mdk_rdev_t, kobj);
1823         kfree(rdev);
1824 }
1825 static struct sysfs_ops rdev_sysfs_ops = {
1826         .show           = rdev_attr_show,
1827         .store          = rdev_attr_store,
1828 };
1829 static struct kobj_type rdev_ktype = {
1830         .release        = rdev_free,
1831         .sysfs_ops      = &rdev_sysfs_ops,
1832         .default_attrs  = rdev_default_attrs,
1833 };
1834
1835 /*
1836  * Import a device. If 'super_format' >= 0, then sanity check the superblock
1837  *
1838  * mark the device faulty if:
1839  *
1840  *   - the device is nonexistent (zero size)
1841  *   - the device has no valid superblock
1842  *
1843  * a faulty rdev _never_ has rdev->sb set.
1844  */
1845 static mdk_rdev_t *md_import_device(dev_t newdev, int super_format, int super_minor)
1846 {
1847         char b[BDEVNAME_SIZE];
1848         int err;
1849         mdk_rdev_t *rdev;
1850         sector_t size;
1851
1852         rdev = kzalloc(sizeof(*rdev), GFP_KERNEL);
1853         if (!rdev) {
1854                 printk(KERN_ERR "md: could not alloc mem for new device!\n");
1855                 return ERR_PTR(-ENOMEM);
1856         }
1857
1858         if ((err = alloc_disk_sb(rdev)))
1859                 goto abort_free;
1860
1861         err = lock_rdev(rdev, newdev);
1862         if (err)
1863                 goto abort_free;
1864
1865         rdev->kobj.parent = NULL;
1866         rdev->kobj.ktype = &rdev_ktype;
1867         kobject_init(&rdev->kobj);
1868
1869         rdev->desc_nr = -1;
1870         rdev->flags = 0;
1871         rdev->data_offset = 0;
1872         atomic_set(&rdev->nr_pending, 0);
1873         atomic_set(&rdev->read_errors, 0);
1874         atomic_set(&rdev->corrected_errors, 0);
1875
1876         size = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
1877         if (!size) {
1878                 printk(KERN_WARNING
1879                         "md: %s has zero or unknown size, marking faulty!\n",
1880                         bdevname(rdev->bdev,b));
1881                 err = -EINVAL;
1882                 goto abort_free;
1883         }
1884
1885         if (super_format >= 0) {
1886                 err = super_types[super_format].
1887                         load_super(rdev, NULL, super_minor);
1888                 if (err == -EINVAL) {
1889                         printk(KERN_WARNING
1890                                 "md: %s has invalid sb, not importing!\n",
1891                                 bdevname(rdev->bdev,b));
1892                         goto abort_free;
1893                 }
1894                 if (err < 0) {
1895                         printk(KERN_WARNING
1896                                 "md: could not read %s's sb, not importing!\n",
1897                                 bdevname(rdev->bdev,b));
1898                         goto abort_free;
1899                 }
1900         }
1901         INIT_LIST_HEAD(&rdev->same_set);
1902
1903         return rdev;
1904
1905 abort_free:
1906         if (rdev->sb_page) {
1907                 if (rdev->bdev)
1908                         unlock_rdev(rdev);
1909                 free_disk_sb(rdev);
1910         }
1911         kfree(rdev);
1912         return ERR_PTR(err);
1913 }
1914
1915 /*
1916  * Check a full RAID array for plausibility
1917  */
1918
1919
1920 static void analyze_sbs(mddev_t * mddev)
1921 {
1922         int i;
1923         struct list_head *tmp;
1924         mdk_rdev_t *rdev, *freshest;
1925         char b[BDEVNAME_SIZE];
1926
1927         freshest = NULL;
1928         ITERATE_RDEV(mddev,rdev,tmp)
1929                 switch (super_types[mddev->major_version].
1930                         load_super(rdev, freshest, mddev->minor_version)) {
1931                 case 1:
1932                         freshest = rdev;
1933                         break;
1934                 case 0:
1935                         break;
1936                 default:
1937                         printk( KERN_ERR \
1938                                 "md: fatal superblock inconsistency in %s"
1939                                 " -- removing from array\n",
1940                                 bdevname(rdev->bdev,b));
1941                         kick_rdev_from_array(rdev);
1942                 }
1943
1944
1945         super_types[mddev->major_version].
1946                 validate_super(mddev, freshest);
1947
1948         i = 0;
1949         ITERATE_RDEV(mddev,rdev,tmp) {
1950                 if (rdev != freshest)
1951                         if (super_types[mddev->major_version].
1952                             validate_super(mddev, rdev)) {
1953                                 printk(KERN_WARNING "md: kicking non-fresh %s"
1954                                         " from array!\n",
1955                                         bdevname(rdev->bdev,b));
1956                                 kick_rdev_from_array(rdev);
1957                                 continue;
1958                         }
1959                 if (mddev->level == LEVEL_MULTIPATH) {
1960                         rdev->desc_nr = i++;
1961                         rdev->raid_disk = rdev->desc_nr;
1962                         set_bit(In_sync, &rdev->flags);
1963                 }
1964         }
1965
1966
1967
1968         if (mddev->recovery_cp != MaxSector &&
1969             mddev->level >= 1)
1970                 printk(KERN_ERR "md: %s: raid array is not clean"
1971                        " -- starting background reconstruction\n",
1972                        mdname(mddev));
1973
1974 }
1975
1976 static ssize_t
1977 level_show(mddev_t *mddev, char *page)
1978 {
1979         struct mdk_personality *p = mddev->pers;
1980         if (p)
1981                 return sprintf(page, "%s\n", p->name);
1982         else if (mddev->clevel[0])
1983                 return sprintf(page, "%s\n", mddev->clevel);
1984         else if (mddev->level != LEVEL_NONE)
1985                 return sprintf(page, "%d\n", mddev->level);
1986         else
1987                 return 0;
1988 }
1989
1990 static ssize_t
1991 level_store(mddev_t *mddev, const char *buf, size_t len)
1992 {
1993         int rv = len;
1994         if (mddev->pers)
1995                 return -EBUSY;
1996         if (len == 0)
1997                 return 0;
1998         if (len >= sizeof(mddev->clevel))
1999                 return -ENOSPC;
2000         strncpy(mddev->clevel, buf, len);
2001         if (mddev->clevel[len-1] == '\n')
2002                 len--;
2003         mddev->clevel[len] = 0;
2004         mddev->level = LEVEL_NONE;
2005         return rv;
2006 }
2007
2008 static struct md_sysfs_entry md_level =
2009 __ATTR(level, 0644, level_show, level_store);
2010
2011 static ssize_t
2012 raid_disks_show(mddev_t *mddev, char *page)
2013 {
2014         if (mddev->raid_disks == 0)
2015                 return 0;
2016         return sprintf(page, "%d\n", mddev->raid_disks);
2017 }
2018
2019 static int update_raid_disks(mddev_t *mddev, int raid_disks);
2020
2021 static ssize_t
2022 raid_disks_store(mddev_t *mddev, const char *buf, size_t len)
2023 {
2024         /* can only set raid_disks if array is not yet active */
2025         char *e;
2026         int rv = 0;
2027         unsigned long n = simple_strtoul(buf, &e, 10);
2028
2029         if (!*buf || (*e && *e != '\n'))
2030                 return -EINVAL;
2031
2032         if (mddev->pers)
2033                 rv = update_raid_disks(mddev, n);
2034         else
2035                 mddev->raid_disks = n;
2036         return rv ? rv : len;
2037 }
2038 static struct md_sysfs_entry md_raid_disks =
2039 __ATTR(raid_disks, 0644, raid_disks_show, raid_disks_store);
2040
2041 static ssize_t
2042 chunk_size_show(mddev_t *mddev, char *page)
2043 {
2044         return sprintf(page, "%d\n", mddev->chunk_size);
2045 }
2046
2047 static ssize_t
2048 chunk_size_store(mddev_t *mddev, const char *buf, size_t len)
2049 {
2050         /* can only set chunk_size if array is not yet active */
2051         char *e;
2052         unsigned long n = simple_strtoul(buf, &e, 10);
2053
2054         if (mddev->pers)
2055                 return -EBUSY;
2056         if (!*buf || (*e && *e != '\n'))
2057                 return -EINVAL;
2058
2059         mddev->chunk_size = n;
2060         return len;
2061 }
2062 static struct md_sysfs_entry md_chunk_size =
2063 __ATTR(chunk_size, 0644, chunk_size_show, chunk_size_store);
2064
2065 static ssize_t
2066 null_show(mddev_t *mddev, char *page)
2067 {
2068         return -EINVAL;
2069 }
2070
2071 static ssize_t
2072 new_dev_store(mddev_t *mddev, const char *buf, size_t len)
2073 {
2074         /* buf must be %d:%d\n? giving major and minor numbers */
2075         /* The new device is added to the array.
2076          * If the array has a persistent superblock, we read the
2077          * superblock to initialise info and check validity.
2078          * Otherwise, only checking done is that in bind_rdev_to_array,
2079          * which mainly checks size.
2080          */
2081         char *e;
2082         int major = simple_strtoul(buf, &e, 10);
2083         int minor;
2084         dev_t dev;
2085         mdk_rdev_t *rdev;
2086         int err;
2087
2088         if (!*buf || *e != ':' || !e[1] || e[1] == '\n')
2089                 return -EINVAL;
2090         minor = simple_strtoul(e+1, &e, 10);
2091         if (*e && *e != '\n')
2092                 return -EINVAL;
2093         dev = MKDEV(major, minor);
2094         if (major != MAJOR(dev) ||
2095             minor != MINOR(dev))
2096                 return -EOVERFLOW;
2097
2098
2099         if (mddev->persistent) {
2100                 rdev = md_import_device(dev, mddev->major_version,
2101                                         mddev->minor_version);
2102                 if (!IS_ERR(rdev) && !list_empty(&mddev->disks)) {
2103                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
2104                                                        mdk_rdev_t, same_set);
2105                         err = super_types[mddev->major_version]
2106                                 .load_super(rdev, rdev0, mddev->minor_version);
2107                         if (err < 0)
2108                                 goto out;
2109                 }
2110         } else
2111                 rdev = md_import_device(dev, -1, -1);
2112
2113         if (IS_ERR(rdev))
2114                 return PTR_ERR(rdev);
2115         err = bind_rdev_to_array(rdev, mddev);
2116  out:
2117         if (err)
2118                 export_rdev(rdev);
2119         return err ? err : len;
2120 }
2121
2122 static struct md_sysfs_entry md_new_device =
2123 __ATTR(new_dev, 0200, null_show, new_dev_store);
2124
2125 static ssize_t
2126 size_show(mddev_t *mddev, char *page)
2127 {
2128         return sprintf(page, "%llu\n", (unsigned long long)mddev->size);
2129 }
2130
2131 static int update_size(mddev_t *mddev, unsigned long size);
2132
2133 static ssize_t
2134 size_store(mddev_t *mddev, const char *buf, size_t len)
2135 {
2136         /* If array is inactive, we can reduce the component size, but
2137          * not increase it (except from 0).
2138          * If array is active, we can try an on-line resize
2139          */
2140         char *e;
2141         int err = 0;
2142         unsigned long long size = simple_strtoull(buf, &e, 10);
2143         if (!*buf || *buf == '\n' ||
2144             (*e && *e != '\n'))
2145                 return -EINVAL;
2146
2147         if (mddev->pers) {
2148                 err = update_size(mddev, size);
2149                 md_update_sb(mddev);
2150         } else {
2151                 if (mddev->size == 0 ||
2152                     mddev->size > size)
2153                         mddev->size = size;
2154                 else
2155                         err = -ENOSPC;
2156         }
2157         return err ? err : len;
2158 }
2159
2160 static struct md_sysfs_entry md_size =
2161 __ATTR(component_size, 0644, size_show, size_store);
2162
2163
2164 /* Metdata version.
2165  * This is either 'none' for arrays with externally managed metadata,
2166  * or N.M for internally known formats
2167  */
2168 static ssize_t
2169 metadata_show(mddev_t *mddev, char *page)
2170 {
2171         if (mddev->persistent)
2172                 return sprintf(page, "%d.%d\n",
2173                                mddev->major_version, mddev->minor_version);
2174         else
2175                 return sprintf(page, "none\n");
2176 }
2177
2178 static ssize_t
2179 metadata_store(mddev_t *mddev, const char *buf, size_t len)
2180 {
2181         int major, minor;
2182         char *e;
2183         if (!list_empty(&mddev->disks))
2184                 return -EBUSY;
2185
2186         if (cmd_match(buf, "none")) {
2187                 mddev->persistent = 0;
2188                 mddev->major_version = 0;
2189                 mddev->minor_version = 90;
2190                 return len;
2191         }
2192         major = simple_strtoul(buf, &e, 10);
2193         if (e==buf || *e != '.')
2194                 return -EINVAL;
2195         buf = e+1;
2196         minor = simple_strtoul(buf, &e, 10);
2197         if (e==buf || *e != '\n')
2198                 return -EINVAL;
2199         if (major >= sizeof(super_types)/sizeof(super_types[0]) ||
2200             super_types[major].name == NULL)
2201                 return -ENOENT;
2202         mddev->major_version = major;
2203         mddev->minor_version = minor;
2204         mddev->persistent = 1;
2205         return len;
2206 }
2207
2208 static struct md_sysfs_entry md_metadata =
2209 __ATTR(metadata_version, 0644, metadata_show, metadata_store);
2210
2211 static ssize_t
2212 action_show(mddev_t *mddev, char *page)
2213 {
2214         char *type = "idle";
2215         if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2216             test_bit(MD_RECOVERY_NEEDED, &mddev->recovery)) {
2217                 if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
2218                         type = "reshape";
2219                 else if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
2220                         if (!test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
2221                                 type = "resync";
2222                         else if (test_bit(MD_RECOVERY_CHECK, &mddev->recovery))
2223                                 type = "check";
2224                         else
2225                                 type = "repair";
2226                 } else
2227                         type = "recover";
2228         }
2229         return sprintf(page, "%s\n", type);
2230 }
2231
2232 static ssize_t
2233 action_store(mddev_t *mddev, const char *page, size_t len)
2234 {
2235         if (!mddev->pers || !mddev->pers->sync_request)
2236                 return -EINVAL;
2237
2238         if (cmd_match(page, "idle")) {
2239                 if (mddev->sync_thread) {
2240                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2241                         md_unregister_thread(mddev->sync_thread);
2242                         mddev->sync_thread = NULL;
2243                         mddev->recovery = 0;
2244                 }
2245         } else if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) ||
2246                    test_bit(MD_RECOVERY_NEEDED, &mddev->recovery))
2247                 return -EBUSY;
2248         else if (cmd_match(page, "resync") || cmd_match(page, "recover"))
2249                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2250         else if (cmd_match(page, "reshape")) {
2251                 int err;
2252                 if (mddev->pers->start_reshape == NULL)
2253                         return -EINVAL;
2254                 err = mddev->pers->start_reshape(mddev);
2255                 if (err)
2256                         return err;
2257         } else {
2258                 if (cmd_match(page, "check"))
2259                         set_bit(MD_RECOVERY_CHECK, &mddev->recovery);
2260                 else if (cmd_match(page, "repair"))
2261                         return -EINVAL;
2262                 set_bit(MD_RECOVERY_REQUESTED, &mddev->recovery);
2263                 set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
2264         }
2265         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2266         md_wakeup_thread(mddev->thread);
2267         return len;
2268 }
2269
2270 static ssize_t
2271 mismatch_cnt_show(mddev_t *mddev, char *page)
2272 {
2273         return sprintf(page, "%llu\n",
2274                        (unsigned long long) mddev->resync_mismatches);
2275 }
2276
2277 static struct md_sysfs_entry
2278 md_scan_mode = __ATTR(sync_action, S_IRUGO|S_IWUSR, action_show, action_store);
2279
2280
2281 static struct md_sysfs_entry
2282 md_mismatches = __ATTR_RO(mismatch_cnt);
2283
2284 static ssize_t
2285 sync_min_show(mddev_t *mddev, char *page)
2286 {
2287         return sprintf(page, "%d (%s)\n", speed_min(mddev),
2288                        mddev->sync_speed_min ? "local": "system");
2289 }
2290
2291 static ssize_t
2292 sync_min_store(mddev_t *mddev, const char *buf, size_t len)
2293 {
2294         int min;
2295         char *e;
2296         if (strncmp(buf, "system", 6)==0) {
2297                 mddev->sync_speed_min = 0;
2298                 return len;
2299         }
2300         min = simple_strtoul(buf, &e, 10);
2301         if (buf == e || (*e && *e != '\n') || min <= 0)
2302                 return -EINVAL;
2303         mddev->sync_speed_min = min;
2304         return len;
2305 }
2306
2307 static struct md_sysfs_entry md_sync_min =
2308 __ATTR(sync_speed_min, S_IRUGO|S_IWUSR, sync_min_show, sync_min_store);
2309
2310 static ssize_t
2311 sync_max_show(mddev_t *mddev, char *page)
2312 {
2313         return sprintf(page, "%d (%s)\n", speed_max(mddev),
2314                        mddev->sync_speed_max ? "local": "system");
2315 }
2316
2317 static ssize_t
2318 sync_max_store(mddev_t *mddev, const char *buf, size_t len)
2319 {
2320         int max;
2321         char *e;
2322         if (strncmp(buf, "system", 6)==0) {
2323                 mddev->sync_speed_max = 0;
2324                 return len;
2325         }
2326         max = simple_strtoul(buf, &e, 10);
2327         if (buf == e || (*e && *e != '\n') || max <= 0)
2328                 return -EINVAL;
2329         mddev->sync_speed_max = max;
2330         return len;
2331 }
2332
2333 static struct md_sysfs_entry md_sync_max =
2334 __ATTR(sync_speed_max, S_IRUGO|S_IWUSR, sync_max_show, sync_max_store);
2335
2336
2337 static ssize_t
2338 sync_speed_show(mddev_t *mddev, char *page)
2339 {
2340         unsigned long resync, dt, db;
2341         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2342         dt = ((jiffies - mddev->resync_mark) / HZ);
2343         if (!dt) dt++;
2344         db = resync - (mddev->resync_mark_cnt);
2345         return sprintf(page, "%ld\n", db/dt/2); /* K/sec */
2346 }
2347
2348 static struct md_sysfs_entry
2349 md_sync_speed = __ATTR_RO(sync_speed);
2350
2351 static ssize_t
2352 sync_completed_show(mddev_t *mddev, char *page)
2353 {
2354         unsigned long max_blocks, resync;
2355
2356         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
2357                 max_blocks = mddev->resync_max_sectors;
2358         else
2359                 max_blocks = mddev->size << 1;
2360
2361         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active));
2362         return sprintf(page, "%lu / %lu\n", resync, max_blocks);
2363 }
2364
2365 static struct md_sysfs_entry
2366 md_sync_completed = __ATTR_RO(sync_completed);
2367
2368 static struct attribute *md_default_attrs[] = {
2369         &md_level.attr,
2370         &md_raid_disks.attr,
2371         &md_chunk_size.attr,
2372         &md_size.attr,
2373         &md_metadata.attr,
2374         &md_new_device.attr,
2375         NULL,
2376 };
2377
2378 static struct attribute *md_redundancy_attrs[] = {
2379         &md_scan_mode.attr,
2380         &md_mismatches.attr,
2381         &md_sync_min.attr,
2382         &md_sync_max.attr,
2383         &md_sync_speed.attr,
2384         &md_sync_completed.attr,
2385         NULL,
2386 };
2387 static struct attribute_group md_redundancy_group = {
2388         .name = NULL,
2389         .attrs = md_redundancy_attrs,
2390 };
2391
2392
2393 static ssize_t
2394 md_attr_show(struct kobject *kobj, struct attribute *attr, char *page)
2395 {
2396         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2397         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2398         ssize_t rv;
2399
2400         if (!entry->show)
2401                 return -EIO;
2402         mddev_lock(mddev);
2403         rv = entry->show(mddev, page);
2404         mddev_unlock(mddev);
2405         return rv;
2406 }
2407
2408 static ssize_t
2409 md_attr_store(struct kobject *kobj, struct attribute *attr,
2410               const char *page, size_t length)
2411 {
2412         struct md_sysfs_entry *entry = container_of(attr, struct md_sysfs_entry, attr);
2413         mddev_t *mddev = container_of(kobj, struct mddev_s, kobj);
2414         ssize_t rv;
2415
2416         if (!entry->store)
2417                 return -EIO;
2418         mddev_lock(mddev);
2419         rv = entry->store(mddev, page, length);
2420         mddev_unlock(mddev);
2421         return rv;
2422 }
2423
2424 static void md_free(struct kobject *ko)
2425 {
2426         mddev_t *mddev = container_of(ko, mddev_t, kobj);
2427         kfree(mddev);
2428 }
2429
2430 static struct sysfs_ops md_sysfs_ops = {
2431         .show   = md_attr_show,
2432         .store  = md_attr_store,
2433 };
2434 static struct kobj_type md_ktype = {
2435         .release        = md_free,
2436         .sysfs_ops      = &md_sysfs_ops,
2437         .default_attrs  = md_default_attrs,
2438 };
2439
2440 int mdp_major = 0;
2441
2442 static struct kobject *md_probe(dev_t dev, int *part, void *data)
2443 {
2444         static DECLARE_MUTEX(disks_sem);
2445         mddev_t *mddev = mddev_find(dev);
2446         struct gendisk *disk;
2447         int partitioned = (MAJOR(dev) != MD_MAJOR);
2448         int shift = partitioned ? MdpMinorShift : 0;
2449         int unit = MINOR(dev) >> shift;
2450
2451         if (!mddev)
2452                 return NULL;
2453
2454         down(&disks_sem);
2455         if (mddev->gendisk) {
2456                 up(&disks_sem);
2457                 mddev_put(mddev);
2458                 return NULL;
2459         }
2460         disk = alloc_disk(1 << shift);
2461         if (!disk) {
2462                 up(&disks_sem);
2463                 mddev_put(mddev);
2464                 return NULL;
2465         }
2466         disk->major = MAJOR(dev);
2467         disk->first_minor = unit << shift;
2468         if (partitioned) {
2469                 sprintf(disk->disk_name, "md_d%d", unit);
2470                 sprintf(disk->devfs_name, "md/d%d", unit);
2471         } else {
2472                 sprintf(disk->disk_name, "md%d", unit);
2473                 sprintf(disk->devfs_name, "md/%d", unit);
2474         }
2475         disk->fops = &md_fops;
2476         disk->private_data = mddev;
2477         disk->queue = mddev->queue;
2478         add_disk(disk);
2479         mddev->gendisk = disk;
2480         up(&disks_sem);
2481         mddev->kobj.parent = &disk->kobj;
2482         mddev->kobj.k_name = NULL;
2483         snprintf(mddev->kobj.name, KOBJ_NAME_LEN, "%s", "md");
2484         mddev->kobj.ktype = &md_ktype;
2485         kobject_register(&mddev->kobj);
2486         return NULL;
2487 }
2488
2489 void md_wakeup_thread(mdk_thread_t *thread);
2490
2491 static void md_safemode_timeout(unsigned long data)
2492 {
2493         mddev_t *mddev = (mddev_t *) data;
2494
2495         mddev->safemode = 1;
2496         md_wakeup_thread(mddev->thread);
2497 }
2498
2499 static int start_dirty_degraded;
2500
2501 static int do_md_run(mddev_t * mddev)
2502 {
2503         int err;
2504         int chunk_size;
2505         struct list_head *tmp;
2506         mdk_rdev_t *rdev;
2507         struct gendisk *disk;
2508         struct mdk_personality *pers;
2509         char b[BDEVNAME_SIZE];
2510
2511         if (list_empty(&mddev->disks))
2512                 /* cannot run an array with no devices.. */
2513                 return -EINVAL;
2514
2515         if (mddev->pers)
2516                 return -EBUSY;
2517
2518         /*
2519          * Analyze all RAID superblock(s)
2520          */
2521         if (!mddev->raid_disks)
2522                 analyze_sbs(mddev);
2523
2524         chunk_size = mddev->chunk_size;
2525
2526         if (chunk_size) {
2527                 if (chunk_size > MAX_CHUNK_SIZE) {
2528                         printk(KERN_ERR "too big chunk_size: %d > %d\n",
2529                                 chunk_size, MAX_CHUNK_SIZE);
2530                         return -EINVAL;
2531                 }
2532                 /*
2533                  * chunk-size has to be a power of 2 and multiples of PAGE_SIZE
2534                  */
2535                 if ( (1 << ffz(~chunk_size)) != chunk_size) {
2536                         printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
2537                         return -EINVAL;
2538                 }
2539                 if (chunk_size < PAGE_SIZE) {
2540                         printk(KERN_ERR "too small chunk_size: %d < %ld\n",
2541                                 chunk_size, PAGE_SIZE);
2542                         return -EINVAL;
2543                 }
2544
2545                 /* devices must have minimum size of one chunk */
2546                 ITERATE_RDEV(mddev,rdev,tmp) {
2547                         if (test_bit(Faulty, &rdev->flags))
2548                                 continue;
2549                         if (rdev->size < chunk_size / 1024) {
2550                                 printk(KERN_WARNING
2551                                         "md: Dev %s smaller than chunk_size:"
2552                                         " %lluk < %dk\n",
2553                                         bdevname(rdev->bdev,b),
2554                                         (unsigned long long)rdev->size,
2555                                         chunk_size / 1024);
2556                                 return -EINVAL;
2557                         }
2558                 }
2559         }
2560
2561 #ifdef CONFIG_KMOD
2562         if (mddev->level != LEVEL_NONE)
2563                 request_module("md-level-%d", mddev->level);
2564         else if (mddev->clevel[0])
2565                 request_module("md-%s", mddev->clevel);
2566 #endif
2567
2568         /*
2569          * Drop all container device buffers, from now on
2570          * the only valid external interface is through the md
2571          * device.
2572          * Also find largest hardsector size
2573          */
2574         ITERATE_RDEV(mddev,rdev,tmp) {
2575                 if (test_bit(Faulty, &rdev->flags))
2576                         continue;
2577                 sync_blockdev(rdev->bdev);
2578                 invalidate_bdev(rdev->bdev, 0);
2579         }
2580
2581         md_probe(mddev->unit, NULL, NULL);
2582         disk = mddev->gendisk;
2583         if (!disk)
2584                 return -ENOMEM;
2585
2586         spin_lock(&pers_lock);
2587         pers = find_pers(mddev->level, mddev->clevel);
2588         if (!pers || !try_module_get(pers->owner)) {
2589                 spin_unlock(&pers_lock);
2590                 if (mddev->level != LEVEL_NONE)
2591                         printk(KERN_WARNING "md: personality for level %d is not loaded!\n",
2592                                mddev->level);
2593                 else
2594                         printk(KERN_WARNING "md: personality for level %s is not loaded!\n",
2595                                mddev->clevel);
2596                 return -EINVAL;
2597         }
2598         mddev->pers = pers;
2599         spin_unlock(&pers_lock);
2600         mddev->level = pers->level;
2601         strlcpy(mddev->clevel, pers->name, sizeof(mddev->clevel));
2602
2603         if (mddev->reshape_position != MaxSector &&
2604             pers->start_reshape == NULL) {
2605                 /* This personality cannot handle reshaping... */
2606                 mddev->pers = NULL;
2607                 module_put(pers->owner);
2608                 return -EINVAL;
2609         }
2610
2611         mddev->recovery = 0;
2612         mddev->resync_max_sectors = mddev->size << 1; /* may be over-ridden by personality */
2613         mddev->barriers_work = 1;
2614         mddev->ok_start_degraded = start_dirty_degraded;
2615
2616         if (start_readonly)
2617                 mddev->ro = 2; /* read-only, but switch on first write */
2618
2619         err = mddev->pers->run(mddev);
2620         if (!err && mddev->pers->sync_request) {
2621                 err = bitmap_create(mddev);
2622                 if (err) {
2623                         printk(KERN_ERR "%s: failed to create bitmap (%d)\n",
2624                                mdname(mddev), err);
2625                         mddev->pers->stop(mddev);
2626                 }
2627         }
2628         if (err) {
2629                 printk(KERN_ERR "md: pers->run() failed ...\n");
2630                 module_put(mddev->pers->owner);
2631                 mddev->pers = NULL;
2632                 bitmap_destroy(mddev);
2633                 return err;
2634         }
2635         if (mddev->pers->sync_request)
2636                 sysfs_create_group(&mddev->kobj, &md_redundancy_group);
2637         else if (mddev->ro == 2) /* auto-readonly not meaningful */
2638                 mddev->ro = 0;
2639
2640         atomic_set(&mddev->writes_pending,0);
2641         mddev->safemode = 0;
2642         mddev->safemode_timer.function = md_safemode_timeout;
2643         mddev->safemode_timer.data = (unsigned long) mddev;
2644         mddev->safemode_delay = (20 * HZ)/1000 +1; /* 20 msec delay */
2645         mddev->in_sync = 1;
2646
2647         ITERATE_RDEV(mddev,rdev,tmp)
2648                 if (rdev->raid_disk >= 0) {
2649                         char nm[20];
2650                         sprintf(nm, "rd%d", rdev->raid_disk);
2651                         sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
2652                 }
2653
2654         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2655         md_wakeup_thread(mddev->thread);
2656
2657         if (mddev->sb_dirty)
2658                 md_update_sb(mddev);
2659
2660         set_capacity(disk, mddev->array_size<<1);
2661
2662         /* If we call blk_queue_make_request here, it will
2663          * re-initialise max_sectors etc which may have been
2664          * refined inside -> run.  So just set the bits we need to set.
2665          * Most initialisation happended when we called
2666          * blk_queue_make_request(..., md_fail_request)
2667          * earlier.
2668          */
2669         mddev->queue->queuedata = mddev;
2670         mddev->queue->make_request_fn = mddev->pers->make_request;
2671
2672         mddev->changed = 1;
2673         md_new_event(mddev);
2674         return 0;
2675 }
2676
2677 static int restart_array(mddev_t *mddev)
2678 {
2679         struct gendisk *disk = mddev->gendisk;
2680         int err;
2681
2682         /*
2683          * Complain if it has no devices
2684          */
2685         err = -ENXIO;
2686         if (list_empty(&mddev->disks))
2687                 goto out;
2688
2689         if (mddev->pers) {
2690                 err = -EBUSY;
2691                 if (!mddev->ro)
2692                         goto out;
2693
2694                 mddev->safemode = 0;
2695                 mddev->ro = 0;
2696                 set_disk_ro(disk, 0);
2697
2698                 printk(KERN_INFO "md: %s switched to read-write mode.\n",
2699                         mdname(mddev));
2700                 /*
2701                  * Kick recovery or resync if necessary
2702                  */
2703                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
2704                 md_wakeup_thread(mddev->thread);
2705                 err = 0;
2706         } else {
2707                 printk(KERN_ERR "md: %s has no personality assigned.\n",
2708                         mdname(mddev));
2709                 err = -EINVAL;
2710         }
2711
2712 out:
2713         return err;
2714 }
2715
2716 static int do_md_stop(mddev_t * mddev, int ro)
2717 {
2718         int err = 0;
2719         struct gendisk *disk = mddev->gendisk;
2720
2721         if (mddev->pers) {
2722                 if (atomic_read(&mddev->active)>2) {
2723                         printk("md: %s still in use.\n",mdname(mddev));
2724                         return -EBUSY;
2725                 }
2726
2727                 if (mddev->sync_thread) {
2728                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
2729                         md_unregister_thread(mddev->sync_thread);
2730                         mddev->sync_thread = NULL;
2731                 }
2732
2733                 del_timer_sync(&mddev->safemode_timer);
2734
2735                 invalidate_partition(disk, 0);
2736
2737                 if (ro) {
2738                         err  = -ENXIO;
2739                         if (mddev->ro==1)
2740                                 goto out;
2741                         mddev->ro = 1;
2742                 } else {
2743                         bitmap_flush(mddev);
2744                         md_super_wait(mddev);
2745                         if (mddev->ro)
2746                                 set_disk_ro(disk, 0);
2747                         blk_queue_make_request(mddev->queue, md_fail_request);
2748                         mddev->pers->stop(mddev);
2749                         if (mddev->pers->sync_request)
2750                                 sysfs_remove_group(&mddev->kobj, &md_redundancy_group);
2751
2752                         module_put(mddev->pers->owner);
2753                         mddev->pers = NULL;
2754                         if (mddev->ro)
2755                                 mddev->ro = 0;
2756                 }
2757                 if (!mddev->in_sync) {
2758                         /* mark array as shutdown cleanly */
2759                         mddev->in_sync = 1;
2760                         md_update_sb(mddev);
2761                 }
2762                 if (ro)
2763                         set_disk_ro(disk, 1);
2764         }
2765
2766         /*
2767          * Free resources if final stop
2768          */
2769         if (!ro) {
2770                 mdk_rdev_t *rdev;
2771                 struct list_head *tmp;
2772                 struct gendisk *disk;
2773                 printk(KERN_INFO "md: %s stopped.\n", mdname(mddev));
2774
2775                 bitmap_destroy(mddev);
2776                 if (mddev->bitmap_file) {
2777                         atomic_set(&mddev->bitmap_file->f_dentry->d_inode->i_writecount, 1);
2778                         fput(mddev->bitmap_file);
2779                         mddev->bitmap_file = NULL;
2780                 }
2781                 mddev->bitmap_offset = 0;
2782
2783                 ITERATE_RDEV(mddev,rdev,tmp)
2784                         if (rdev->raid_disk >= 0) {
2785                                 char nm[20];
2786                                 sprintf(nm, "rd%d", rdev->raid_disk);
2787                                 sysfs_remove_link(&mddev->kobj, nm);
2788                         }
2789
2790                 export_array(mddev);
2791
2792                 mddev->array_size = 0;
2793                 disk = mddev->gendisk;
2794                 if (disk)
2795                         set_capacity(disk, 0);
2796                 mddev->changed = 1;
2797         } else
2798                 printk(KERN_INFO "md: %s switched to read-only mode.\n",
2799                         mdname(mddev));
2800         err = 0;
2801         md_new_event(mddev);
2802 out:
2803         return err;
2804 }
2805
2806 static void autorun_array(mddev_t *mddev)
2807 {
2808         mdk_rdev_t *rdev;
2809         struct list_head *tmp;
2810         int err;
2811
2812         if (list_empty(&mddev->disks))
2813                 return;
2814
2815         printk(KERN_INFO "md: running: ");
2816
2817         ITERATE_RDEV(mddev,rdev,tmp) {
2818                 char b[BDEVNAME_SIZE];
2819                 printk("<%s>", bdevname(rdev->bdev,b));
2820         }
2821         printk("\n");
2822
2823         err = do_md_run (mddev);
2824         if (err) {
2825                 printk(KERN_WARNING "md: do_md_run() returned %d\n", err);
2826                 do_md_stop (mddev, 0);
2827         }
2828 }
2829
2830 /*
2831  * lets try to run arrays based on all disks that have arrived
2832  * until now. (those are in pending_raid_disks)
2833  *
2834  * the method: pick the first pending disk, collect all disks with
2835  * the same UUID, remove all from the pending list and put them into
2836  * the 'same_array' list. Then order this list based on superblock
2837  * update time (freshest comes first), kick out 'old' disks and
2838  * compare superblocks. If everything's fine then run it.
2839  *
2840  * If "unit" is allocated, then bump its reference count
2841  */
2842 static void autorun_devices(int part)
2843 {
2844         struct list_head *tmp;
2845         mdk_rdev_t *rdev0, *rdev;
2846         mddev_t *mddev;
2847         char b[BDEVNAME_SIZE];
2848
2849         printk(KERN_INFO "md: autorun ...\n");
2850         while (!list_empty(&pending_raid_disks)) {
2851                 dev_t dev;
2852                 LIST_HEAD(candidates);
2853                 rdev0 = list_entry(pending_raid_disks.next,
2854                                          mdk_rdev_t, same_set);
2855
2856                 printk(KERN_INFO "md: considering %s ...\n",
2857                         bdevname(rdev0->bdev,b));
2858                 INIT_LIST_HEAD(&candidates);
2859                 ITERATE_RDEV_PENDING(rdev,tmp)
2860                         if (super_90_load(rdev, rdev0, 0) >= 0) {
2861                                 printk(KERN_INFO "md:  adding %s ...\n",
2862                                         bdevname(rdev->bdev,b));
2863                                 list_move(&rdev->same_set, &candidates);
2864                         }
2865                 /*
2866                  * now we have a set of devices, with all of them having
2867                  * mostly sane superblocks. It's time to allocate the
2868                  * mddev.
2869                  */
2870                 if (rdev0->preferred_minor < 0 || rdev0->preferred_minor >= MAX_MD_DEVS) {
2871                         printk(KERN_INFO "md: unit number in %s is bad: %d\n",
2872                                bdevname(rdev0->bdev, b), rdev0->preferred_minor);
2873                         break;
2874                 }
2875                 if (part)
2876                         dev = MKDEV(mdp_major,
2877                                     rdev0->preferred_minor << MdpMinorShift);
2878                 else
2879                         dev = MKDEV(MD_MAJOR, rdev0->preferred_minor);
2880
2881                 md_probe(dev, NULL, NULL);
2882                 mddev = mddev_find(dev);
2883                 if (!mddev) {
2884                         printk(KERN_ERR
2885                                 "md: cannot allocate memory for md drive.\n");
2886                         break;
2887                 }
2888                 if (mddev_lock(mddev))
2889                         printk(KERN_WARNING "md: %s locked, cannot run\n",
2890                                mdname(mddev));
2891                 else if (mddev->raid_disks || mddev->major_version
2892                          || !list_empty(&mddev->disks)) {
2893                         printk(KERN_WARNING
2894                                 "md: %s already running, cannot run %s\n",
2895                                 mdname(mddev), bdevname(rdev0->bdev,b));
2896                         mddev_unlock(mddev);
2897                 } else {
2898                         printk(KERN_INFO "md: created %s\n", mdname(mddev));
2899                         ITERATE_RDEV_GENERIC(candidates,rdev,tmp) {
2900                                 list_del_init(&rdev->same_set);
2901                                 if (bind_rdev_to_array(rdev, mddev))
2902                                         export_rdev(rdev);
2903                         }
2904                         autorun_array(mddev);
2905                         mddev_unlock(mddev);
2906                 }
2907                 /* on success, candidates will be empty, on error
2908                  * it won't...
2909                  */
2910                 ITERATE_RDEV_GENERIC(candidates,rdev,tmp)
2911                         export_rdev(rdev);
2912                 mddev_put(mddev);
2913         }
2914         printk(KERN_INFO "md: ... autorun DONE.\n");
2915 }
2916
2917 /*
2918  * import RAID devices based on one partition
2919  * if possible, the array gets run as well.
2920  */
2921
2922 static int autostart_array(dev_t startdev)
2923 {
2924         char b[BDEVNAME_SIZE];
2925         int err = -EINVAL, i;
2926         mdp_super_t *sb = NULL;
2927         mdk_rdev_t *start_rdev = NULL, *rdev;
2928
2929         start_rdev = md_import_device(startdev, 0, 0);
2930         if (IS_ERR(start_rdev))
2931                 return err;
2932
2933
2934         /* NOTE: this can only work for 0.90.0 superblocks */
2935         sb = (mdp_super_t*)page_address(start_rdev->sb_page);
2936         if (sb->major_version != 0 ||
2937             sb->minor_version != 90 ) {
2938                 printk(KERN_WARNING "md: can only autostart 0.90.0 arrays\n");
2939                 export_rdev(start_rdev);
2940                 return err;
2941         }
2942
2943         if (test_bit(Faulty, &start_rdev->flags)) {
2944                 printk(KERN_WARNING
2945                         "md: can not autostart based on faulty %s!\n",
2946                         bdevname(start_rdev->bdev,b));
2947                 export_rdev(start_rdev);
2948                 return err;
2949         }
2950         list_add(&start_rdev->same_set, &pending_raid_disks);
2951
2952         for (i = 0; i < MD_SB_DISKS; i++) {
2953                 mdp_disk_t *desc = sb->disks + i;
2954                 dev_t dev = MKDEV(desc->major, desc->minor);
2955
2956                 if (!dev)
2957                         continue;
2958                 if (dev == startdev)
2959                         continue;
2960                 if (MAJOR(dev) != desc->major || MINOR(dev) != desc->minor)
2961                         continue;
2962                 rdev = md_import_device(dev, 0, 0);
2963                 if (IS_ERR(rdev))
2964                         continue;
2965
2966                 list_add(&rdev->same_set, &pending_raid_disks);
2967         }
2968
2969         /*
2970          * possibly return codes
2971          */
2972         autorun_devices(0);
2973         return 0;
2974
2975 }
2976
2977
2978 static int get_version(void __user * arg)
2979 {
2980         mdu_version_t ver;
2981
2982         ver.major = MD_MAJOR_VERSION;
2983         ver.minor = MD_MINOR_VERSION;
2984         ver.patchlevel = MD_PATCHLEVEL_VERSION;
2985
2986         if (copy_to_user(arg, &ver, sizeof(ver)))
2987                 return -EFAULT;
2988
2989         return 0;
2990 }
2991
2992 static int get_array_info(mddev_t * mddev, void __user * arg)
2993 {
2994         mdu_array_info_t info;
2995         int nr,working,active,failed,spare;
2996         mdk_rdev_t *rdev;
2997         struct list_head *tmp;
2998
2999         nr=working=active=failed=spare=0;
3000         ITERATE_RDEV(mddev,rdev,tmp) {
3001                 nr++;
3002                 if (test_bit(Faulty, &rdev->flags))
3003                         failed++;
3004                 else {
3005                         working++;
3006                         if (test_bit(In_sync, &rdev->flags))
3007                                 active++;
3008                         else
3009                                 spare++;
3010                 }
3011         }
3012
3013         info.major_version = mddev->major_version;
3014         info.minor_version = mddev->minor_version;
3015         info.patch_version = MD_PATCHLEVEL_VERSION;
3016         info.ctime         = mddev->ctime;
3017         info.level         = mddev->level;
3018         info.size          = mddev->size;
3019         if (info.size != mddev->size) /* overflow */
3020                 info.size = -1;
3021         info.nr_disks      = nr;
3022         info.raid_disks    = mddev->raid_disks;
3023         info.md_minor      = mddev->md_minor;
3024         info.not_persistent= !mddev->persistent;
3025
3026         info.utime         = mddev->utime;
3027         info.state         = 0;
3028         if (mddev->in_sync)
3029                 info.state = (1<<MD_SB_CLEAN);
3030         if (mddev->bitmap && mddev->bitmap_offset)
3031                 info.state = (1<<MD_SB_BITMAP_PRESENT);
3032         info.active_disks  = active;
3033         info.working_disks = working;
3034         info.failed_disks  = failed;
3035         info.spare_disks   = spare;
3036
3037         info.layout        = mddev->layout;
3038         info.chunk_size    = mddev->chunk_size;
3039
3040         if (copy_to_user(arg, &info, sizeof(info)))
3041                 return -EFAULT;
3042
3043         return 0;
3044 }
3045
3046 static int get_bitmap_file(mddev_t * mddev, void __user * arg)
3047 {
3048         mdu_bitmap_file_t *file = NULL; /* too big for stack allocation */
3049         char *ptr, *buf = NULL;
3050         int err = -ENOMEM;
3051
3052         file = kmalloc(sizeof(*file), GFP_KERNEL);
3053         if (!file)
3054                 goto out;
3055
3056         /* bitmap disabled, zero the first byte and copy out */
3057         if (!mddev->bitmap || !mddev->bitmap->file) {
3058                 file->pathname[0] = '\0';
3059                 goto copy_out;
3060         }
3061
3062         buf = kmalloc(sizeof(file->pathname), GFP_KERNEL);
3063         if (!buf)
3064                 goto out;
3065
3066         ptr = file_path(mddev->bitmap->file, buf, sizeof(file->pathname));
3067         if (!ptr)
3068                 goto out;
3069
3070         strcpy(file->pathname, ptr);
3071
3072 copy_out:
3073         err = 0;
3074         if (copy_to_user(arg, file, sizeof(*file)))
3075                 err = -EFAULT;
3076 out:
3077         kfree(buf);
3078         kfree(file);
3079         return err;
3080 }
3081
3082 static int get_disk_info(mddev_t * mddev, void __user * arg)
3083 {
3084         mdu_disk_info_t info;
3085         unsigned int nr;
3086         mdk_rdev_t *rdev;
3087
3088         if (copy_from_user(&info, arg, sizeof(info)))
3089                 return -EFAULT;
3090
3091         nr = info.number;
3092
3093         rdev = find_rdev_nr(mddev, nr);
3094         if (rdev) {
3095                 info.major = MAJOR(rdev->bdev->bd_dev);
3096                 info.minor = MINOR(rdev->bdev->bd_dev);
3097                 info.raid_disk = rdev->raid_disk;
3098                 info.state = 0;
3099                 if (test_bit(Faulty, &rdev->flags))
3100                         info.state |= (1<<MD_DISK_FAULTY);
3101                 else if (test_bit(In_sync, &rdev->flags)) {
3102                         info.state |= (1<<MD_DISK_ACTIVE);
3103                         info.state |= (1<<MD_DISK_SYNC);
3104                 }
3105                 if (test_bit(WriteMostly, &rdev->flags))
3106                         info.state |= (1<<MD_DISK_WRITEMOSTLY);
3107         } else {
3108                 info.major = info.minor = 0;
3109                 info.raid_disk = -1;
3110                 info.state = (1<<MD_DISK_REMOVED);
3111         }
3112
3113         if (copy_to_user(arg, &info, sizeof(info)))
3114                 return -EFAULT;
3115
3116         return 0;
3117 }
3118
3119 static int add_new_disk(mddev_t * mddev, mdu_disk_info_t *info)
3120 {
3121         char b[BDEVNAME_SIZE], b2[BDEVNAME_SIZE];
3122         mdk_rdev_t *rdev;
3123         dev_t dev = MKDEV(info->major,info->minor);
3124
3125         if (info->major != MAJOR(dev) || info->minor != MINOR(dev))
3126                 return -EOVERFLOW;
3127
3128         if (!mddev->raid_disks) {
3129                 int err;
3130                 /* expecting a device which has a superblock */
3131                 rdev = md_import_device(dev, mddev->major_version, mddev->minor_version);
3132                 if (IS_ERR(rdev)) {
3133                         printk(KERN_WARNING
3134                                 "md: md_import_device returned %ld\n",
3135                                 PTR_ERR(rdev));
3136                         return PTR_ERR(rdev);
3137                 }
3138                 if (!list_empty(&mddev->disks)) {
3139                         mdk_rdev_t *rdev0 = list_entry(mddev->disks.next,
3140                                                         mdk_rdev_t, same_set);
3141                         int err = super_types[mddev->major_version]
3142                                 .load_super(rdev, rdev0, mddev->minor_version);
3143                         if (err < 0) {
3144                                 printk(KERN_WARNING
3145                                         "md: %s has different UUID to %s\n",
3146                                         bdevname(rdev->bdev,b),
3147                                         bdevname(rdev0->bdev,b2));
3148                                 export_rdev(rdev);
3149                                 return -EINVAL;
3150                         }
3151                 }
3152                 err = bind_rdev_to_array(rdev, mddev);
3153                 if (err)
3154                         export_rdev(rdev);
3155                 return err;
3156         }
3157
3158         /*
3159          * add_new_disk can be used once the array is assembled
3160          * to add "hot spares".  They must already have a superblock
3161          * written
3162          */
3163         if (mddev->pers) {
3164                 int err;
3165                 if (!mddev->pers->hot_add_disk) {
3166                         printk(KERN_WARNING
3167                                 "%s: personality does not support diskops!\n",
3168                                mdname(mddev));
3169                         return -EINVAL;
3170                 }
3171                 if (mddev->persistent)
3172                         rdev = md_import_device(dev, mddev->major_version,
3173                                                 mddev->minor_version);
3174                 else
3175                         rdev = md_import_device(dev, -1, -1);
3176                 if (IS_ERR(rdev)) {
3177                         printk(KERN_WARNING
3178                                 "md: md_import_device returned %ld\n",
3179                                 PTR_ERR(rdev));
3180                         return PTR_ERR(rdev);
3181                 }
3182                 /* set save_raid_disk if appropriate */
3183                 if (!mddev->persistent) {
3184                         if (info->state & (1<<MD_DISK_SYNC)  &&
3185                             info->raid_disk < mddev->raid_disks)
3186                                 rdev->raid_disk = info->raid_disk;
3187                         else
3188                                 rdev->raid_disk = -1;
3189                 } else
3190                         super_types[mddev->major_version].
3191                                 validate_super(mddev, rdev);
3192                 rdev->saved_raid_disk = rdev->raid_disk;
3193
3194                 clear_bit(In_sync, &rdev->flags); /* just to be sure */
3195                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3196                         set_bit(WriteMostly, &rdev->flags);
3197
3198                 rdev->raid_disk = -1;
3199                 err = bind_rdev_to_array(rdev, mddev);
3200                 if (err)
3201                         export_rdev(rdev);
3202
3203                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3204                 md_wakeup_thread(mddev->thread);
3205                 return err;
3206         }
3207
3208         /* otherwise, add_new_disk is only allowed
3209          * for major_version==0 superblocks
3210          */
3211         if (mddev->major_version != 0) {
3212                 printk(KERN_WARNING "%s: ADD_NEW_DISK not supported\n",
3213                        mdname(mddev));
3214                 return -EINVAL;
3215         }
3216
3217         if (!(info->state & (1<<MD_DISK_FAULTY))) {
3218                 int err;
3219                 rdev = md_import_device (dev, -1, 0);
3220                 if (IS_ERR(rdev)) {
3221                         printk(KERN_WARNING
3222                                 "md: error, md_import_device() returned %ld\n",
3223                                 PTR_ERR(rdev));
3224                         return PTR_ERR(rdev);
3225                 }
3226                 rdev->desc_nr = info->number;
3227                 if (info->raid_disk < mddev->raid_disks)
3228                         rdev->raid_disk = info->raid_disk;
3229                 else
3230                         rdev->raid_disk = -1;
3231
3232                 rdev->flags = 0;
3233
3234                 if (rdev->raid_disk < mddev->raid_disks)
3235                         if (info->state & (1<<MD_DISK_SYNC))
3236                                 set_bit(In_sync, &rdev->flags);
3237
3238                 if (info->state & (1<<MD_DISK_WRITEMOSTLY))
3239                         set_bit(WriteMostly, &rdev->flags);
3240
3241                 if (!mddev->persistent) {
3242                         printk(KERN_INFO "md: nonpersistent superblock ...\n");
3243                         rdev->sb_offset = rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3244                 } else
3245                         rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3246                 rdev->size = calc_dev_size(rdev, mddev->chunk_size);
3247
3248                 err = bind_rdev_to_array(rdev, mddev);
3249                 if (err) {
3250                         export_rdev(rdev);
3251                         return err;
3252                 }
3253         }
3254
3255         return 0;
3256 }
3257
3258 static int hot_remove_disk(mddev_t * mddev, dev_t dev)
3259 {
3260         char b[BDEVNAME_SIZE];
3261         mdk_rdev_t *rdev;
3262
3263         if (!mddev->pers)
3264                 return -ENODEV;
3265
3266         rdev = find_rdev(mddev, dev);
3267         if (!rdev)
3268                 return -ENXIO;
3269
3270         if (rdev->raid_disk >= 0)
3271                 goto busy;
3272
3273         kick_rdev_from_array(rdev);
3274         md_update_sb(mddev);
3275         md_new_event(mddev);
3276
3277         return 0;
3278 busy:
3279         printk(KERN_WARNING "md: cannot remove active disk %s from %s ... \n",
3280                 bdevname(rdev->bdev,b), mdname(mddev));
3281         return -EBUSY;
3282 }
3283
3284 static int hot_add_disk(mddev_t * mddev, dev_t dev)
3285 {
3286         char b[BDEVNAME_SIZE];
3287         int err;
3288         unsigned int size;
3289         mdk_rdev_t *rdev;
3290
3291         if (!mddev->pers)
3292                 return -ENODEV;
3293
3294         if (mddev->major_version != 0) {
3295                 printk(KERN_WARNING "%s: HOT_ADD may only be used with"
3296                         " version-0 superblocks.\n",
3297                         mdname(mddev));
3298                 return -EINVAL;
3299         }
3300         if (!mddev->pers->hot_add_disk) {
3301                 printk(KERN_WARNING
3302                         "%s: personality does not support diskops!\n",
3303                         mdname(mddev));
3304                 return -EINVAL;
3305         }
3306
3307         rdev = md_import_device (dev, -1, 0);
3308         if (IS_ERR(rdev)) {
3309                 printk(KERN_WARNING
3310                         "md: error, md_import_device() returned %ld\n",
3311                         PTR_ERR(rdev));
3312                 return -EINVAL;
3313         }
3314
3315         if (mddev->persistent)
3316                 rdev->sb_offset = calc_dev_sboffset(rdev->bdev);
3317         else
3318                 rdev->sb_offset =
3319                         rdev->bdev->bd_inode->i_size >> BLOCK_SIZE_BITS;
3320
3321         size = calc_dev_size(rdev, mddev->chunk_size);
3322         rdev->size = size;
3323
3324         if (test_bit(Faulty, &rdev->flags)) {
3325                 printk(KERN_WARNING
3326                         "md: can not hot-add faulty %s disk to %s!\n",
3327                         bdevname(rdev->bdev,b), mdname(mddev));
3328                 err = -EINVAL;
3329                 goto abort_export;
3330         }
3331         clear_bit(In_sync, &rdev->flags);
3332         rdev->desc_nr = -1;
3333         err = bind_rdev_to_array(rdev, mddev);
3334         if (err)
3335                 goto abort_export;
3336
3337         /*
3338          * The rest should better be atomic, we can have disk failures
3339          * noticed in interrupt contexts ...
3340          */
3341
3342         if (rdev->desc_nr == mddev->max_disks) {
3343                 printk(KERN_WARNING "%s: can not hot-add to full array!\n",
3344                         mdname(mddev));
3345                 err = -EBUSY;
3346                 goto abort_unbind_export;
3347         }
3348
3349         rdev->raid_disk = -1;
3350
3351         md_update_sb(mddev);
3352
3353         /*
3354          * Kick recovery, maybe this spare has to be added to the
3355          * array immediately.
3356          */
3357         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3358         md_wakeup_thread(mddev->thread);
3359         md_new_event(mddev);
3360         return 0;
3361
3362 abort_unbind_export:
3363         unbind_rdev_from_array(rdev);
3364
3365 abort_export:
3366         export_rdev(rdev);
3367         return err;
3368 }
3369
3370 /* similar to deny_write_access, but accounts for our holding a reference
3371  * to the file ourselves */
3372 static int deny_bitmap_write_access(struct file * file)
3373 {
3374         struct inode *inode = file->f_mapping->host;
3375
3376         spin_lock(&inode->i_lock);
3377         if (atomic_read(&inode->i_writecount) > 1) {
3378                 spin_unlock(&inode->i_lock);
3379                 return -ETXTBSY;
3380         }
3381         atomic_set(&inode->i_writecount, -1);
3382         spin_unlock(&inode->i_lock);
3383
3384         return 0;
3385 }
3386
3387 static int set_bitmap_file(mddev_t *mddev, int fd)
3388 {
3389         int err;
3390
3391         if (mddev->pers) {
3392                 if (!mddev->pers->quiesce)
3393                         return -EBUSY;
3394                 if (mddev->recovery || mddev->sync_thread)
3395                         return -EBUSY;
3396                 /* we should be able to change the bitmap.. */
3397         }
3398
3399
3400         if (fd >= 0) {
3401                 if (mddev->bitmap)
3402                         return -EEXIST; /* cannot add when bitmap is present */
3403                 mddev->bitmap_file = fget(fd);
3404
3405                 if (mddev->bitmap_file == NULL) {
3406                         printk(KERN_ERR "%s: error: failed to get bitmap file\n",
3407                                mdname(mddev));
3408                         return -EBADF;
3409                 }
3410
3411                 err = deny_bitmap_write_access(mddev->bitmap_file);
3412                 if (err) {
3413                         printk(KERN_ERR "%s: error: bitmap file is already in use\n",
3414                                mdname(mddev));
3415                         fput(mddev->bitmap_file);
3416                         mddev->bitmap_file = NULL;
3417                         return err;
3418                 }
3419                 mddev->bitmap_offset = 0; /* file overrides offset */
3420         } else if (mddev->bitmap == NULL)
3421                 return -ENOENT; /* cannot remove what isn't there */
3422         err = 0;
3423         if (mddev->pers) {
3424                 mddev->pers->quiesce(mddev, 1);
3425                 if (fd >= 0)
3426                         err = bitmap_create(mddev);
3427                 if (fd < 0 || err)
3428                         bitmap_destroy(mddev);
3429                 mddev->pers->quiesce(mddev, 0);
3430         } else if (fd < 0) {
3431                 if (mddev->bitmap_file)
3432                         fput(mddev->bitmap_file);
3433                 mddev->bitmap_file = NULL;
3434         }
3435
3436         return err;
3437 }
3438
3439 /*
3440  * set_array_info is used two different ways
3441  * The original usage is when creating a new array.
3442  * In this usage, raid_disks is > 0 and it together with
3443  *  level, size, not_persistent,layout,chunksize determine the
3444  *  shape of the array.
3445  *  This will always create an array with a type-0.90.0 superblock.
3446  * The newer usage is when assembling an array.
3447  *  In this case raid_disks will be 0, and the major_version field is
3448  *  use to determine which style super-blocks are to be found on the devices.
3449  *  The minor and patch _version numbers are also kept incase the
3450  *  super_block handler wishes to interpret them.
3451  */
3452 static int set_array_info(mddev_t * mddev, mdu_array_info_t *info)
3453 {
3454
3455         if (info->raid_disks == 0) {
3456                 /* just setting version number for superblock loading */
3457                 if (info->major_version < 0 ||
3458                     info->major_version >= sizeof(super_types)/sizeof(super_types[0]) ||
3459                     super_types[info->major_version].name == NULL) {
3460                         /* maybe try to auto-load a module? */
3461                         printk(KERN_INFO
3462                                 "md: superblock version %d not known\n",
3463                                 info->major_version);
3464                         return -EINVAL;
3465                 }
3466                 mddev->major_version = info->major_version;
3467                 mddev->minor_version = info->minor_version;
3468                 mddev->patch_version = info->patch_version;
3469                 return 0;
3470         }
3471         mddev->major_version = MD_MAJOR_VERSION;
3472         mddev->minor_version = MD_MINOR_VERSION;
3473         mddev->patch_version = MD_PATCHLEVEL_VERSION;
3474         mddev->ctime         = get_seconds();
3475
3476         mddev->level         = info->level;
3477         mddev->clevel[0]     = 0;
3478         mddev->size          = info->size;
3479         mddev->raid_disks    = info->raid_disks;
3480         /* don't set md_minor, it is determined by which /dev/md* was
3481          * openned
3482          */
3483         if (info->state & (1<<MD_SB_CLEAN))
3484                 mddev->recovery_cp = MaxSector;
3485         else
3486                 mddev->recovery_cp = 0;
3487         mddev->persistent    = ! info->not_persistent;
3488
3489         mddev->layout        = info->layout;
3490         mddev->chunk_size    = info->chunk_size;
3491
3492         mddev->max_disks     = MD_SB_DISKS;
3493
3494         mddev->sb_dirty      = 1;
3495
3496         mddev->default_bitmap_offset = MD_SB_BYTES >> 9;
3497         mddev->bitmap_offset = 0;
3498
3499         mddev->reshape_position = MaxSector;
3500
3501         /*
3502          * Generate a 128 bit UUID
3503          */
3504         get_random_bytes(mddev->uuid, 16);
3505
3506         mddev->new_level = mddev->level;
3507         mddev->new_chunk = mddev->chunk_size;
3508         mddev->new_layout = mddev->layout;
3509         mddev->delta_disks = 0;
3510
3511         return 0;
3512 }
3513
3514 static int update_size(mddev_t *mddev, unsigned long size)
3515 {
3516         mdk_rdev_t * rdev;
3517         int rv;
3518         struct list_head *tmp;
3519
3520         if (mddev->pers->resize == NULL)
3521                 return -EINVAL;
3522         /* The "size" is the amount of each device that is used.
3523          * This can only make sense for arrays with redundancy.
3524          * linear and raid0 always use whatever space is available
3525          * We can only consider changing the size if no resync
3526          * or reconstruction is happening, and if the new size
3527          * is acceptable. It must fit before the sb_offset or,
3528          * if that is <data_offset, it must fit before the
3529          * size of each device.
3530          * If size is zero, we find the largest size that fits.
3531          */
3532         if (mddev->sync_thread)
3533                 return -EBUSY;
3534         ITERATE_RDEV(mddev,rdev,tmp) {
3535                 sector_t avail;
3536                 int fit = (size == 0);
3537                 if (rdev->sb_offset > rdev->data_offset)
3538                         avail = (rdev->sb_offset*2) - rdev->data_offset;
3539                 else
3540                         avail = get_capacity(rdev->bdev->bd_disk)
3541                                 - rdev->data_offset;
3542                 if (fit && (size == 0 || size > avail/2))
3543                         size = avail/2;
3544                 if (avail < ((sector_t)size << 1))
3545                         return -ENOSPC;
3546         }
3547         rv = mddev->pers->resize(mddev, (sector_t)size *2);
3548         if (!rv) {
3549                 struct block_device *bdev;
3550
3551                 bdev = bdget_disk(mddev->gendisk, 0);
3552                 if (bdev) {
3553                         mutex_lock(&bdev->bd_inode->i_mutex);
3554                         i_size_write(bdev->bd_inode, (loff_t)mddev->array_size << 10);
3555                         mutex_unlock(&bdev->bd_inode->i_mutex);
3556                         bdput(bdev);
3557                 }
3558         }
3559         return rv;
3560 }
3561
3562 static int update_raid_disks(mddev_t *mddev, int raid_disks)
3563 {
3564         int rv;
3565         /* change the number of raid disks */
3566         if (mddev->pers->check_reshape == NULL)
3567                 return -EINVAL;
3568         if (raid_disks <= 0 ||
3569             raid_disks >= mddev->max_disks)
3570                 return -EINVAL;
3571         if (mddev->sync_thread || mddev->reshape_position != MaxSector)
3572                 return -EBUSY;
3573         mddev->delta_disks = raid_disks - mddev->raid_disks;
3574
3575         rv = mddev->pers->check_reshape(mddev);
3576         return rv;
3577 }
3578
3579
3580 /*
3581  * update_array_info is used to change the configuration of an
3582  * on-line array.
3583  * The version, ctime,level,size,raid_disks,not_persistent, layout,chunk_size
3584  * fields in the info are checked against the array.
3585  * Any differences that cannot be handled will cause an error.
3586  * Normally, only one change can be managed at a time.
3587  */
3588 static int update_array_info(mddev_t *mddev, mdu_array_info_t *info)
3589 {
3590         int rv = 0;
3591         int cnt = 0;
3592         int state = 0;
3593
3594         /* calculate expected state,ignoring low bits */
3595         if (mddev->bitmap && mddev->bitmap_offset)
3596                 state |= (1 << MD_SB_BITMAP_PRESENT);
3597
3598         if (mddev->major_version != info->major_version ||
3599             mddev->minor_version != info->minor_version ||
3600 /*          mddev->patch_version != info->patch_version || */
3601             mddev->ctime         != info->ctime         ||
3602             mddev->level         != info->level         ||
3603 /*          mddev->layout        != info->layout        || */
3604             !mddev->persistent   != info->not_persistent||
3605             mddev->chunk_size    != info->chunk_size    ||
3606             /* ignore bottom 8 bits of state, and allow SB_BITMAP_PRESENT to change */
3607             ((state^info->state) & 0xfffffe00)
3608                 )
3609                 return -EINVAL;
3610         /* Check there is only one change */
3611         if (info->size >= 0 && mddev->size != info->size) cnt++;
3612         if (mddev->raid_disks != info->raid_disks) cnt++;
3613         if (mddev->layout != info->layout) cnt++;
3614         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) cnt++;
3615         if (cnt == 0) return 0;
3616         if (cnt > 1) return -EINVAL;
3617
3618         if (mddev->layout != info->layout) {
3619                 /* Change layout
3620                  * we don't need to do anything at the md level, the
3621                  * personality will take care of it all.
3622                  */
3623                 if (mddev->pers->reconfig == NULL)
3624                         return -EINVAL;
3625                 else
3626                         return mddev->pers->reconfig(mddev, info->layout, -1);
3627         }
3628         if (info->size >= 0 && mddev->size != info->size)
3629                 rv = update_size(mddev, info->size);
3630
3631         if (mddev->raid_disks    != info->raid_disks)
3632                 rv = update_raid_disks(mddev, info->raid_disks);
3633
3634         if ((state ^ info->state) & (1<<MD_SB_BITMAP_PRESENT)) {
3635                 if (mddev->pers->quiesce == NULL)
3636                         return -EINVAL;
3637                 if (mddev->recovery || mddev->sync_thread)
3638                         return -EBUSY;
3639                 if (info->state & (1<<MD_SB_BITMAP_PRESENT)) {
3640                         /* add the bitmap */
3641                         if (mddev->bitmap)
3642                                 return -EEXIST;
3643                         if (mddev->default_bitmap_offset == 0)
3644                                 return -EINVAL;
3645                         mddev->bitmap_offset = mddev->default_bitmap_offset;
3646                         mddev->pers->quiesce(mddev, 1);
3647                         rv = bitmap_create(mddev);
3648                         if (rv)
3649                                 bitmap_destroy(mddev);
3650                         mddev->pers->quiesce(mddev, 0);
3651                 } else {
3652                         /* remove the bitmap */
3653                         if (!mddev->bitmap)
3654                                 return -ENOENT;
3655                         if (mddev->bitmap->file)
3656                                 return -EINVAL;
3657                         mddev->pers->quiesce(mddev, 1);
3658                         bitmap_destroy(mddev);
3659                         mddev->pers->quiesce(mddev, 0);
3660                         mddev->bitmap_offset = 0;
3661                 }
3662         }
3663         md_update_sb(mddev);
3664         return rv;
3665 }
3666
3667 static int set_disk_faulty(mddev_t *mddev, dev_t dev)
3668 {
3669         mdk_rdev_t *rdev;
3670
3671         if (mddev->pers == NULL)
3672                 return -ENODEV;
3673
3674         rdev = find_rdev(mddev, dev);
3675         if (!rdev)
3676                 return -ENODEV;
3677
3678         md_error(mddev, rdev);
3679         return 0;
3680 }
3681
3682 static int md_getgeo(struct block_device *bdev, struct hd_geometry *geo)
3683 {
3684         mddev_t *mddev = bdev->bd_disk->private_data;
3685
3686         geo->heads = 2;
3687         geo->sectors = 4;
3688         geo->cylinders = get_capacity(mddev->gendisk) / 8;
3689         return 0;
3690 }
3691
3692 static int md_ioctl(struct inode *inode, struct file *file,
3693                         unsigned int cmd, unsigned long arg)
3694 {
3695         int err = 0;
3696         void __user *argp = (void __user *)arg;
3697         mddev_t *mddev = NULL;
3698
3699         if (!capable(CAP_SYS_ADMIN))
3700                 return -EACCES;
3701
3702         /*
3703          * Commands dealing with the RAID driver but not any
3704          * particular array:
3705          */
3706         switch (cmd)
3707         {
3708                 case RAID_VERSION:
3709                         err = get_version(argp);
3710                         goto done;
3711
3712                 case PRINT_RAID_DEBUG:
3713                         err = 0;
3714                         md_print_devices();
3715                         goto done;
3716
3717 #ifndef MODULE
3718                 case RAID_AUTORUN:
3719                         err = 0;
3720                         autostart_arrays(arg);
3721                         goto done;
3722 #endif
3723                 default:;
3724         }
3725
3726         /*
3727          * Commands creating/starting a new array:
3728          */
3729
3730         mddev = inode->i_bdev->bd_disk->private_data;
3731
3732         if (!mddev) {
3733                 BUG();
3734                 goto abort;
3735         }
3736
3737
3738         if (cmd == START_ARRAY) {
3739                 /* START_ARRAY doesn't need to lock the array as autostart_array
3740                  * does the locking, and it could even be a different array
3741                  */
3742                 static int cnt = 3;
3743                 if (cnt > 0 ) {
3744                         printk(KERN_WARNING
3745                                "md: %s(pid %d) used deprecated START_ARRAY ioctl. "
3746                                "This will not be supported beyond July 2006\n",
3747                                current->comm, current->pid);
3748                         cnt--;
3749                 }
3750                 err = autostart_array(new_decode_dev(arg));
3751                 if (err) {
3752                         printk(KERN_WARNING "md: autostart failed!\n");
3753                         goto abort;
3754                 }
3755                 goto done;
3756         }
3757
3758         err = mddev_lock(mddev);
3759         if (err) {
3760                 printk(KERN_INFO
3761                         "md: ioctl lock interrupted, reason %d, cmd %d\n",
3762                         err, cmd);
3763                 goto abort;
3764         }
3765
3766         switch (cmd)
3767         {
3768                 case SET_ARRAY_INFO:
3769                         {
3770                                 mdu_array_info_t info;
3771                                 if (!arg)
3772                                         memset(&info, 0, sizeof(info));
3773                                 else if (copy_from_user(&info, argp, sizeof(info))) {
3774                                         err = -EFAULT;
3775                                         goto abort_unlock;
3776                                 }
3777                                 if (mddev->pers) {
3778                                         err = update_array_info(mddev, &info);
3779                                         if (err) {
3780                                                 printk(KERN_WARNING "md: couldn't update"
3781                                                        " array info. %d\n", err);
3782                                                 goto abort_unlock;
3783                                         }
3784                                         goto done_unlock;
3785                                 }
3786                                 if (!list_empty(&mddev->disks)) {
3787                                         printk(KERN_WARNING
3788                                                "md: array %s already has disks!\n",
3789                                                mdname(mddev));
3790                                         err = -EBUSY;
3791                                         goto abort_unlock;
3792                                 }
3793                                 if (mddev->raid_disks) {
3794                                         printk(KERN_WARNING
3795                                                "md: array %s already initialised!\n",
3796                                                mdname(mddev));
3797                                         err = -EBUSY;
3798                                         goto abort_unlock;
3799                                 }
3800                                 err = set_array_info(mddev, &info);
3801                                 if (err) {
3802                                         printk(KERN_WARNING "md: couldn't set"
3803                                                " array info. %d\n", err);
3804                                         goto abort_unlock;
3805                                 }
3806                         }
3807                         goto done_unlock;
3808
3809                 default:;
3810         }
3811
3812         /*
3813          * Commands querying/configuring an existing array:
3814          */
3815         /* if we are not initialised yet, only ADD_NEW_DISK, STOP_ARRAY,
3816          * RUN_ARRAY, and SET_BITMAP_FILE are allowed */
3817         if (!mddev->raid_disks && cmd != ADD_NEW_DISK && cmd != STOP_ARRAY
3818                         && cmd != RUN_ARRAY && cmd != SET_BITMAP_FILE) {
3819                 err = -ENODEV;
3820                 goto abort_unlock;
3821         }
3822
3823         /*
3824          * Commands even a read-only array can execute:
3825          */
3826         switch (cmd)
3827         {
3828                 case GET_ARRAY_INFO:
3829                         err = get_array_info(mddev, argp);
3830                         goto done_unlock;
3831
3832                 case GET_BITMAP_FILE:
3833                         err = get_bitmap_file(mddev, argp);
3834                         goto done_unlock;
3835
3836                 case GET_DISK_INFO:
3837                         err = get_disk_info(mddev, argp);
3838                         goto done_unlock;
3839
3840                 case RESTART_ARRAY_RW:
3841                         err = restart_array(mddev);
3842                         goto done_unlock;
3843
3844                 case STOP_ARRAY:
3845                         err = do_md_stop (mddev, 0);
3846                         goto done_unlock;
3847
3848                 case STOP_ARRAY_RO:
3849                         err = do_md_stop (mddev, 1);
3850                         goto done_unlock;
3851
3852         /*
3853          * We have a problem here : there is no easy way to give a CHS
3854          * virtual geometry. We currently pretend that we have a 2 heads
3855          * 4 sectors (with a BIG number of cylinders...). This drives
3856          * dosfs just mad... ;-)
3857          */
3858         }
3859
3860         /*
3861          * The remaining ioctls are changing the state of the
3862          * superblock, so we do not allow them on read-only arrays.
3863          * However non-MD ioctls (e.g. get-size) will still come through
3864          * here and hit the 'default' below, so only disallow
3865          * 'md' ioctls, and switch to rw mode if started auto-readonly.
3866          */
3867         if (_IOC_TYPE(cmd) == MD_MAJOR &&
3868             mddev->ro && mddev->pers) {
3869                 if (mddev->ro == 2) {
3870                         mddev->ro = 0;
3871                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
3872                 md_wakeup_thread(mddev->thread);
3873
3874                 } else {
3875                         err = -EROFS;
3876                         goto abort_unlock;
3877                 }
3878         }
3879
3880         switch (cmd)
3881         {
3882                 case ADD_NEW_DISK:
3883                 {
3884                         mdu_disk_info_t info;
3885                         if (copy_from_user(&info, argp, sizeof(info)))
3886                                 err = -EFAULT;
3887                         else
3888                                 err = add_new_disk(mddev, &info);
3889                         goto done_unlock;
3890                 }
3891
3892                 case HOT_REMOVE_DISK:
3893                         err = hot_remove_disk(mddev, new_decode_dev(arg));
3894                         goto done_unlock;
3895
3896                 case HOT_ADD_DISK:
3897                         err = hot_add_disk(mddev, new_decode_dev(arg));
3898                         goto done_unlock;
3899
3900                 case SET_DISK_FAULTY:
3901                         err = set_disk_faulty(mddev, new_decode_dev(arg));
3902                         goto done_unlock;
3903
3904                 case RUN_ARRAY:
3905                         err = do_md_run (mddev);
3906                         goto done_unlock;
3907
3908                 case SET_BITMAP_FILE:
3909                         err = set_bitmap_file(mddev, (int)arg);
3910                         goto done_unlock;
3911
3912                 default:
3913                         if (_IOC_TYPE(cmd) == MD_MAJOR)
3914                                 printk(KERN_WARNING "md: %s(pid %d) used"
3915                                         " obsolete MD ioctl, upgrade your"
3916                                         " software to use new ictls.\n",
3917                                         current->comm, current->pid);
3918                         err = -EINVAL;
3919                         goto abort_unlock;
3920         }
3921
3922 done_unlock:
3923 abort_unlock:
3924         mddev_unlock(mddev);
3925
3926         return err;
3927 done:
3928         if (err)
3929                 MD_BUG();
3930 abort:
3931         return err;
3932 }
3933
3934 static int md_open(struct inode *inode, struct file *file)
3935 {
3936         /*
3937          * Succeed if we can lock the mddev, which confirms that
3938          * it isn't being stopped right now.
3939          */
3940         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3941         int err;
3942
3943         if ((err = mddev_lock(mddev)))
3944                 goto out;
3945
3946         err = 0;
3947         mddev_get(mddev);
3948         mddev_unlock(mddev);
3949
3950         check_disk_change(inode->i_bdev);
3951  out:
3952         return err;
3953 }
3954
3955 static int md_release(struct inode *inode, struct file * file)
3956 {
3957         mddev_t *mddev = inode->i_bdev->bd_disk->private_data;
3958
3959         if (!mddev)
3960                 BUG();
3961         mddev_put(mddev);
3962
3963         return 0;
3964 }
3965
3966 static int md_media_changed(struct gendisk *disk)
3967 {
3968         mddev_t *mddev = disk->private_data;
3969
3970         return mddev->changed;
3971 }
3972
3973 static int md_revalidate(struct gendisk *disk)
3974 {
3975         mddev_t *mddev = disk->private_data;
3976
3977         mddev->changed = 0;
3978         return 0;
3979 }
3980 static struct block_device_operations md_fops =
3981 {
3982         .owner          = THIS_MODULE,
3983         .open           = md_open,
3984         .release        = md_release,
3985         .ioctl          = md_ioctl,
3986         .getgeo         = md_getgeo,
3987         .media_changed  = md_media_changed,
3988         .revalidate_disk= md_revalidate,
3989 };
3990
3991 static int md_thread(void * arg)
3992 {
3993         mdk_thread_t *thread = arg;
3994
3995         /*
3996          * md_thread is a 'system-thread', it's priority should be very
3997          * high. We avoid resource deadlocks individually in each
3998          * raid personality. (RAID5 does preallocation) We also use RR and
3999          * the very same RT priority as kswapd, thus we will never get
4000          * into a priority inversion deadlock.
4001          *
4002          * we definitely have to have equal or higher priority than
4003          * bdflush, otherwise bdflush will deadlock if there are too
4004          * many dirty RAID5 blocks.
4005          */
4006
4007         allow_signal(SIGKILL);
4008         while (!kthread_should_stop()) {
4009
4010                 /* We need to wait INTERRUPTIBLE so that
4011                  * we don't add to the load-average.
4012                  * That means we need to be sure no signals are
4013                  * pending
4014                  */
4015                 if (signal_pending(current))
4016                         flush_signals(current);
4017
4018                 wait_event_interruptible_timeout
4019                         (thread->wqueue,
4020                          test_bit(THREAD_WAKEUP, &thread->flags)
4021                          || kthread_should_stop(),
4022                          thread->timeout);
4023                 try_to_freeze();
4024
4025                 clear_bit(THREAD_WAKEUP, &thread->flags);
4026
4027                 thread->run(thread->mddev);
4028         }
4029
4030         return 0;
4031 }
4032
4033 void md_wakeup_thread(mdk_thread_t *thread)
4034 {
4035         if (thread) {
4036                 dprintk("md: waking up MD thread %s.\n", thread->tsk->comm);
4037                 set_bit(THREAD_WAKEUP, &thread->flags);
4038                 wake_up(&thread->wqueue);
4039         }
4040 }
4041
4042 mdk_thread_t *md_register_thread(void (*run) (mddev_t *), mddev_t *mddev,
4043                                  const char *name)
4044 {
4045         mdk_thread_t *thread;
4046
4047         thread = kzalloc(sizeof(mdk_thread_t), GFP_KERNEL);
4048         if (!thread)
4049                 return NULL;
4050
4051         init_waitqueue_head(&thread->wqueue);
4052
4053         thread->run = run;
4054         thread->mddev = mddev;
4055         thread->timeout = MAX_SCHEDULE_TIMEOUT;
4056         thread->tsk = kthread_run(md_thread, thread, name, mdname(thread->mddev));
4057         if (IS_ERR(thread->tsk)) {
4058                 kfree(thread);
4059                 return NULL;
4060         }
4061         return thread;
4062 }
4063
4064 void md_unregister_thread(mdk_thread_t *thread)
4065 {
4066         dprintk("interrupting MD-thread pid %d\n", thread->tsk->pid);
4067
4068         kthread_stop(thread->tsk);
4069         kfree(thread);
4070 }
4071
4072 void md_error(mddev_t *mddev, mdk_rdev_t *rdev)
4073 {
4074         if (!mddev) {
4075                 MD_BUG();
4076                 return;
4077         }
4078
4079         if (!rdev || test_bit(Faulty, &rdev->flags))
4080                 return;
4081 /*
4082         dprintk("md_error dev:%s, rdev:(%d:%d), (caller: %p,%p,%p,%p).\n",
4083                 mdname(mddev),
4084                 MAJOR(rdev->bdev->bd_dev), MINOR(rdev->bdev->bd_dev),
4085                 __builtin_return_address(0),__builtin_return_address(1),
4086                 __builtin_return_address(2),__builtin_return_address(3));
4087 */
4088         if (!mddev->pers->error_handler)
4089                 return;
4090         mddev->pers->error_handler(mddev,rdev);
4091         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4092         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4093         md_wakeup_thread(mddev->thread);
4094         md_new_event(mddev);
4095 }
4096
4097 /* seq_file implementation /proc/mdstat */
4098
4099 static void status_unused(struct seq_file *seq)
4100 {
4101         int i = 0;
4102         mdk_rdev_t *rdev;
4103         struct list_head *tmp;
4104
4105         seq_printf(seq, "unused devices: ");
4106
4107         ITERATE_RDEV_PENDING(rdev,tmp) {
4108                 char b[BDEVNAME_SIZE];
4109                 i++;
4110                 seq_printf(seq, "%s ",
4111                               bdevname(rdev->bdev,b));
4112         }
4113         if (!i)
4114                 seq_printf(seq, "<none>");
4115
4116         seq_printf(seq, "\n");
4117 }
4118
4119
4120 static void status_resync(struct seq_file *seq, mddev_t * mddev)
4121 {
4122         sector_t max_blocks, resync, res;
4123         unsigned long dt, db, rt;
4124         int scale;
4125         unsigned int per_milli;
4126
4127         resync = (mddev->curr_resync - atomic_read(&mddev->recovery_active))/2;
4128
4129         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4130                 max_blocks = mddev->resync_max_sectors >> 1;
4131         else
4132                 max_blocks = mddev->size;
4133
4134         /*
4135          * Should not happen.
4136          */
4137         if (!max_blocks) {
4138                 MD_BUG();
4139                 return;
4140         }
4141         /* Pick 'scale' such that (resync>>scale)*1000 will fit
4142          * in a sector_t, and (max_blocks>>scale) will fit in a
4143          * u32, as those are the requirements for sector_div.
4144          * Thus 'scale' must be at least 10
4145          */
4146         scale = 10;
4147         if (sizeof(sector_t) > sizeof(unsigned long)) {
4148                 while ( max_blocks/2 > (1ULL<<(scale+32)))
4149                         scale++;
4150         }
4151         res = (resync>>scale)*1000;
4152         sector_div(res, (u32)((max_blocks>>scale)+1));
4153
4154         per_milli = res;
4155         {
4156                 int i, x = per_milli/50, y = 20-x;
4157                 seq_printf(seq, "[");
4158                 for (i = 0; i < x; i++)
4159                         seq_printf(seq, "=");
4160                 seq_printf(seq, ">");
4161                 for (i = 0; i < y; i++)
4162                         seq_printf(seq, ".");
4163                 seq_printf(seq, "] ");
4164         }
4165         seq_printf(seq, " %s =%3u.%u%% (%llu/%llu)",
4166                    (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery)?
4167                     "reshape" :
4168                       (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) ?
4169                        "resync" : "recovery")),
4170                       per_milli/10, per_milli % 10,
4171                    (unsigned long long) resync,
4172                    (unsigned long long) max_blocks);
4173
4174         /*
4175          * We do not want to overflow, so the order of operands and
4176          * the * 100 / 100 trick are important. We do a +1 to be
4177          * safe against division by zero. We only estimate anyway.
4178          *
4179          * dt: time from mark until now
4180          * db: blocks written from mark until now
4181          * rt: remaining time
4182          */
4183         dt = ((jiffies - mddev->resync_mark) / HZ);
4184         if (!dt) dt++;
4185         db = resync - (mddev->resync_mark_cnt/2);
4186         rt = (dt * ((unsigned long)(max_blocks-resync) / (db/100+1)))/100;
4187
4188         seq_printf(seq, " finish=%lu.%lumin", rt / 60, (rt % 60)/6);
4189
4190         seq_printf(seq, " speed=%ldK/sec", db/dt);
4191 }
4192
4193 static void *md_seq_start(struct seq_file *seq, loff_t *pos)
4194 {
4195         struct list_head *tmp;
4196         loff_t l = *pos;
4197         mddev_t *mddev;
4198
4199         if (l >= 0x10000)
4200                 return NULL;
4201         if (!l--)
4202                 /* header */
4203                 return (void*)1;
4204
4205         spin_lock(&all_mddevs_lock);
4206         list_for_each(tmp,&all_mddevs)
4207                 if (!l--) {
4208                         mddev = list_entry(tmp, mddev_t, all_mddevs);
4209                         mddev_get(mddev);
4210                         spin_unlock(&all_mddevs_lock);
4211                         return mddev;
4212                 }
4213         spin_unlock(&all_mddevs_lock);
4214         if (!l--)
4215                 return (void*)2;/* tail */
4216         return NULL;
4217 }
4218
4219 static void *md_seq_next(struct seq_file *seq, void *v, loff_t *pos)
4220 {
4221         struct list_head *tmp;
4222         mddev_t *next_mddev, *mddev = v;
4223
4224         ++*pos;
4225         if (v == (void*)2)
4226                 return NULL;
4227
4228         spin_lock(&all_mddevs_lock);
4229         if (v == (void*)1)
4230                 tmp = all_mddevs.next;
4231         else
4232                 tmp = mddev->all_mddevs.next;
4233         if (tmp != &all_mddevs)
4234                 next_mddev = mddev_get(list_entry(tmp,mddev_t,all_mddevs));
4235         else {
4236                 next_mddev = (void*)2;
4237                 *pos = 0x10000;
4238         }
4239         spin_unlock(&all_mddevs_lock);
4240
4241         if (v != (void*)1)
4242                 mddev_put(mddev);
4243         return next_mddev;
4244
4245 }
4246
4247 static void md_seq_stop(struct seq_file *seq, void *v)
4248 {
4249         mddev_t *mddev = v;
4250
4251         if (mddev && v != (void*)1 && v != (void*)2)
4252                 mddev_put(mddev);
4253 }
4254
4255 struct mdstat_info {
4256         int event;
4257 };
4258
4259 static int md_seq_show(struct seq_file *seq, void *v)
4260 {
4261         mddev_t *mddev = v;
4262         sector_t size;
4263         struct list_head *tmp2;
4264         mdk_rdev_t *rdev;
4265         struct mdstat_info *mi = seq->private;
4266         struct bitmap *bitmap;
4267
4268         if (v == (void*)1) {
4269                 struct mdk_personality *pers;
4270                 seq_printf(seq, "Personalities : ");
4271                 spin_lock(&pers_lock);
4272                 list_for_each_entry(pers, &pers_list, list)
4273                         seq_printf(seq, "[%s] ", pers->name);
4274
4275                 spin_unlock(&pers_lock);
4276                 seq_printf(seq, "\n");
4277                 mi->event = atomic_read(&md_event_count);
4278                 return 0;
4279         }
4280         if (v == (void*)2) {
4281                 status_unused(seq);
4282                 return 0;
4283         }
4284
4285         if (mddev_lock(mddev)!=0)
4286                 return -EINTR;
4287         if (mddev->pers || mddev->raid_disks || !list_empty(&mddev->disks)) {
4288                 seq_printf(seq, "%s : %sactive", mdname(mddev),
4289                                                 mddev->pers ? "" : "in");
4290                 if (mddev->pers) {
4291                         if (mddev->ro==1)
4292                                 seq_printf(seq, " (read-only)");
4293                         if (mddev->ro==2)
4294                                 seq_printf(seq, "(auto-read-only)");
4295                         seq_printf(seq, " %s", mddev->pers->name);
4296                 }
4297
4298                 size = 0;
4299                 ITERATE_RDEV(mddev,rdev,tmp2) {
4300                         char b[BDEVNAME_SIZE];
4301                         seq_printf(seq, " %s[%d]",
4302                                 bdevname(rdev->bdev,b), rdev->desc_nr);
4303                         if (test_bit(WriteMostly, &rdev->flags))
4304                                 seq_printf(seq, "(W)");
4305                         if (test_bit(Faulty, &rdev->flags)) {
4306                                 seq_printf(seq, "(F)");
4307                                 continue;
4308                         } else if (rdev->raid_disk < 0)
4309                                 seq_printf(seq, "(S)"); /* spare */
4310                         size += rdev->size;
4311                 }
4312
4313                 if (!list_empty(&mddev->disks)) {
4314                         if (mddev->pers)
4315                                 seq_printf(seq, "\n      %llu blocks",
4316                                         (unsigned long long)mddev->array_size);
4317                         else
4318                                 seq_printf(seq, "\n      %llu blocks",
4319                                         (unsigned long long)size);
4320                 }
4321                 if (mddev->persistent) {
4322                         if (mddev->major_version != 0 ||
4323                             mddev->minor_version != 90) {
4324                                 seq_printf(seq," super %d.%d",
4325                                            mddev->major_version,
4326                                            mddev->minor_version);
4327                         }
4328                 } else
4329                         seq_printf(seq, " super non-persistent");
4330
4331                 if (mddev->pers) {
4332                         mddev->pers->status (seq, mddev);
4333                         seq_printf(seq, "\n      ");
4334                         if (mddev->pers->sync_request) {
4335                                 if (mddev->curr_resync > 2) {
4336                                         status_resync (seq, mddev);
4337                                         seq_printf(seq, "\n      ");
4338                                 } else if (mddev->curr_resync == 1 || mddev->curr_resync == 2)
4339                                         seq_printf(seq, "\tresync=DELAYED\n      ");
4340                                 else if (mddev->recovery_cp < MaxSector)
4341                                         seq_printf(seq, "\tresync=PENDING\n      ");
4342                         }
4343                 } else
4344                         seq_printf(seq, "\n       ");
4345
4346                 if ((bitmap = mddev->bitmap)) {
4347                         unsigned long chunk_kb;
4348                         unsigned long flags;
4349                         spin_lock_irqsave(&bitmap->lock, flags);
4350                         chunk_kb = bitmap->chunksize >> 10;
4351                         seq_printf(seq, "bitmap: %lu/%lu pages [%luKB], "
4352                                 "%lu%s chunk",
4353                                 bitmap->pages - bitmap->missing_pages,
4354                                 bitmap->pages,
4355                                 (bitmap->pages - bitmap->missing_pages)
4356                                         << (PAGE_SHIFT - 10),
4357                                 chunk_kb ? chunk_kb : bitmap->chunksize,
4358                                 chunk_kb ? "KB" : "B");
4359                         if (bitmap->file) {
4360                                 seq_printf(seq, ", file: ");
4361                                 seq_path(seq, bitmap->file->f_vfsmnt,
4362                                          bitmap->file->f_dentry," \t\n");
4363                         }
4364
4365                         seq_printf(seq, "\n");
4366                         spin_unlock_irqrestore(&bitmap->lock, flags);
4367                 }
4368
4369                 seq_printf(seq, "\n");
4370         }
4371         mddev_unlock(mddev);
4372
4373         return 0;
4374 }
4375
4376 static struct seq_operations md_seq_ops = {
4377         .start  = md_seq_start,
4378         .next   = md_seq_next,
4379         .stop   = md_seq_stop,
4380         .show   = md_seq_show,
4381 };
4382
4383 static int md_seq_open(struct inode *inode, struct file *file)
4384 {
4385         int error;
4386         struct mdstat_info *mi = kmalloc(sizeof(*mi), GFP_KERNEL);
4387         if (mi == NULL)
4388                 return -ENOMEM;
4389
4390         error = seq_open(file, &md_seq_ops);
4391         if (error)
4392                 kfree(mi);
4393         else {
4394                 struct seq_file *p = file->private_data;
4395                 p->private = mi;
4396                 mi->event = atomic_read(&md_event_count);
4397         }
4398         return error;
4399 }
4400
4401 static int md_seq_release(struct inode *inode, struct file *file)
4402 {
4403         struct seq_file *m = file->private_data;
4404         struct mdstat_info *mi = m->private;
4405         m->private = NULL;
4406         kfree(mi);
4407         return seq_release(inode, file);
4408 }
4409
4410 static unsigned int mdstat_poll(struct file *filp, poll_table *wait)
4411 {
4412         struct seq_file *m = filp->private_data;
4413         struct mdstat_info *mi = m->private;
4414         int mask;
4415
4416         poll_wait(filp, &md_event_waiters, wait);
4417
4418         /* always allow read */
4419         mask = POLLIN | POLLRDNORM;
4420
4421         if (mi->event != atomic_read(&md_event_count))
4422                 mask |= POLLERR | POLLPRI;
4423         return mask;
4424 }
4425
4426 static struct file_operations md_seq_fops = {
4427         .open           = md_seq_open,
4428         .read           = seq_read,
4429         .llseek         = seq_lseek,
4430         .release        = md_seq_release,
4431         .poll           = mdstat_poll,
4432 };
4433
4434 int register_md_personality(struct mdk_personality *p)
4435 {
4436         spin_lock(&pers_lock);
4437         list_add_tail(&p->list, &pers_list);
4438         printk(KERN_INFO "md: %s personality registered for level %d\n", p->name, p->level);
4439         spin_unlock(&pers_lock);
4440         return 0;
4441 }
4442
4443 int unregister_md_personality(struct mdk_personality *p)
4444 {
4445         printk(KERN_INFO "md: %s personality unregistered\n", p->name);
4446         spin_lock(&pers_lock);
4447         list_del_init(&p->list);
4448         spin_unlock(&pers_lock);
4449         return 0;
4450 }
4451
4452 static int is_mddev_idle(mddev_t *mddev)
4453 {
4454         mdk_rdev_t * rdev;
4455         struct list_head *tmp;
4456         int idle;
4457         unsigned long curr_events;
4458
4459         idle = 1;
4460         ITERATE_RDEV(mddev,rdev,tmp) {
4461                 struct gendisk *disk = rdev->bdev->bd_contains->bd_disk;
4462                 curr_events = disk_stat_read(disk, sectors[0]) +
4463                                 disk_stat_read(disk, sectors[1]) -
4464                                 atomic_read(&disk->sync_io);
4465                 /* The difference between curr_events and last_events
4466                  * will be affected by any new non-sync IO (making
4467                  * curr_events bigger) and any difference in the amount of
4468                  * in-flight syncio (making current_events bigger or smaller)
4469                  * The amount in-flight is currently limited to
4470                  * 32*64K in raid1/10 and 256*PAGE_SIZE in raid5/6
4471                  * which is at most 4096 sectors.
4472                  * These numbers are fairly fragile and should be made
4473                  * more robust, probably by enforcing the
4474                  * 'window size' that md_do_sync sort-of uses.
4475                  *
4476                  * Note: the following is an unsigned comparison.
4477                  */
4478                 if ((curr_events - rdev->last_events + 4096) > 8192) {
4479                         rdev->last_events = curr_events;
4480                         idle = 0;
4481                 }
4482         }
4483         return idle;
4484 }
4485
4486 void md_done_sync(mddev_t *mddev, int blocks, int ok)
4487 {
4488         /* another "blocks" (512byte) blocks have been synced */
4489         atomic_sub(blocks, &mddev->recovery_active);
4490         wake_up(&mddev->recovery_wait);
4491         if (!ok) {
4492                 set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4493                 md_wakeup_thread(mddev->thread);
4494                 // stop recovery, signal do_sync ....
4495         }
4496 }
4497
4498
4499 /* md_write_start(mddev, bi)
4500  * If we need to update some array metadata (e.g. 'active' flag
4501  * in superblock) before writing, schedule a superblock update
4502  * and wait for it to complete.
4503  */
4504 void md_write_start(mddev_t *mddev, struct bio *bi)
4505 {
4506         if (bio_data_dir(bi) != WRITE)
4507                 return;
4508
4509         BUG_ON(mddev->ro == 1);
4510         if (mddev->ro == 2) {
4511                 /* need to switch to read/write */
4512                 mddev->ro = 0;
4513                 set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4514                 md_wakeup_thread(mddev->thread);
4515         }
4516         atomic_inc(&mddev->writes_pending);
4517         if (mddev->in_sync) {
4518                 spin_lock_irq(&mddev->write_lock);
4519                 if (mddev->in_sync) {
4520                         mddev->in_sync = 0;
4521                         mddev->sb_dirty = 1;
4522                         md_wakeup_thread(mddev->thread);
4523                 }
4524                 spin_unlock_irq(&mddev->write_lock);
4525         }
4526         wait_event(mddev->sb_wait, mddev->sb_dirty==0);
4527 }
4528
4529 void md_write_end(mddev_t *mddev)
4530 {
4531         if (atomic_dec_and_test(&mddev->writes_pending)) {
4532                 if (mddev->safemode == 2)
4533                         md_wakeup_thread(mddev->thread);
4534                 else
4535                         mod_timer(&mddev->safemode_timer, jiffies + mddev->safemode_delay);
4536         }
4537 }
4538
4539 static DECLARE_WAIT_QUEUE_HEAD(resync_wait);
4540
4541 #define SYNC_MARKS      10
4542 #define SYNC_MARK_STEP  (3*HZ)
4543 void md_do_sync(mddev_t *mddev)
4544 {
4545         mddev_t *mddev2;
4546         unsigned int currspeed = 0,
4547                  window;
4548         sector_t max_sectors,j, io_sectors;
4549         unsigned long mark[SYNC_MARKS];
4550         sector_t mark_cnt[SYNC_MARKS];
4551         int last_mark,m;
4552         struct list_head *tmp;
4553         sector_t last_check;
4554         int skipped = 0;
4555
4556         /* just incase thread restarts... */
4557         if (test_bit(MD_RECOVERY_DONE, &mddev->recovery))
4558                 return;
4559
4560         /* we overload curr_resync somewhat here.
4561          * 0 == not engaged in resync at all
4562          * 2 == checking that there is no conflict with another sync
4563          * 1 == like 2, but have yielded to allow conflicting resync to
4564          *              commense
4565          * other == active in resync - this many blocks
4566          *
4567          * Before starting a resync we must have set curr_resync to
4568          * 2, and then checked that every "conflicting" array has curr_resync
4569          * less than ours.  When we find one that is the same or higher
4570          * we wait on resync_wait.  To avoid deadlock, we reduce curr_resync
4571          * to 1 if we choose to yield (based arbitrarily on address of mddev structure).
4572          * This will mean we have to start checking from the beginning again.
4573          *
4574          */
4575
4576         do {
4577                 mddev->curr_resync = 2;
4578
4579         try_again:
4580                 if (kthread_should_stop()) {
4581                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4582                         goto skip;
4583                 }
4584                 ITERATE_MDDEV(mddev2,tmp) {
4585                         if (mddev2 == mddev)
4586                                 continue;
4587                         if (mddev2->curr_resync &&
4588                             match_mddev_units(mddev,mddev2)) {
4589                                 DEFINE_WAIT(wq);
4590                                 if (mddev < mddev2 && mddev->curr_resync == 2) {
4591                                         /* arbitrarily yield */
4592                                         mddev->curr_resync = 1;
4593                                         wake_up(&resync_wait);
4594                                 }
4595                                 if (mddev > mddev2 && mddev->curr_resync == 1)
4596                                         /* no need to wait here, we can wait the next
4597                                          * time 'round when curr_resync == 2
4598                                          */
4599                                         continue;
4600                                 prepare_to_wait(&resync_wait, &wq, TASK_UNINTERRUPTIBLE);
4601                                 if (!kthread_should_stop() &&
4602                                     mddev2->curr_resync >= mddev->curr_resync) {
4603                                         printk(KERN_INFO "md: delaying resync of %s"
4604                                                " until %s has finished resync (they"
4605                                                " share one or more physical units)\n",
4606                                                mdname(mddev), mdname(mddev2));
4607                                         mddev_put(mddev2);
4608                                         schedule();
4609                                         finish_wait(&resync_wait, &wq);
4610                                         goto try_again;
4611                                 }
4612                                 finish_wait(&resync_wait, &wq);
4613                         }
4614                 }
4615         } while (mddev->curr_resync < 2);
4616
4617         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery)) {
4618                 /* resync follows the size requested by the personality,
4619                  * which defaults to physical size, but can be virtual size
4620                  */
4621                 max_sectors = mddev->resync_max_sectors;
4622                 mddev->resync_mismatches = 0;
4623         } else if (test_bit(MD_RECOVERY_RESHAPE, &mddev->recovery))
4624                 max_sectors = mddev->size << 1;
4625         else
4626                 /* recovery follows the physical size of devices */
4627                 max_sectors = mddev->size << 1;
4628
4629         printk(KERN_INFO "md: syncing RAID array %s\n", mdname(mddev));
4630         printk(KERN_INFO "md: minimum _guaranteed_ reconstruction speed:"
4631                 " %d KB/sec/disc.\n", speed_min(mddev));
4632         printk(KERN_INFO "md: using maximum available idle IO bandwidth "
4633                "(but not more than %d KB/sec) for reconstruction.\n",
4634                speed_max(mddev));
4635
4636         is_mddev_idle(mddev); /* this also initializes IO event counters */
4637         /* we don't use the checkpoint if there's a bitmap */
4638         if (test_bit(MD_RECOVERY_SYNC, &mddev->recovery) && !mddev->bitmap
4639             && ! test_bit(MD_RECOVERY_REQUESTED, &mddev->recovery))
4640                 j = mddev->recovery_cp;
4641         else
4642                 j = 0;
4643         io_sectors = 0;
4644         for (m = 0; m < SYNC_MARKS; m++) {
4645                 mark[m] = jiffies;
4646                 mark_cnt[m] = io_sectors;
4647         }
4648         last_mark = 0;
4649         mddev->resync_mark = mark[last_mark];
4650         mddev->resync_mark_cnt = mark_cnt[last_mark];
4651
4652         /*
4653          * Tune reconstruction:
4654          */
4655         window = 32*(PAGE_SIZE/512);
4656         printk(KERN_INFO "md: using %dk window, over a total of %llu blocks.\n",
4657                 window/2,(unsigned long long) max_sectors/2);
4658
4659         atomic_set(&mddev->recovery_active, 0);
4660         init_waitqueue_head(&mddev->recovery_wait);
4661         last_check = 0;
4662
4663         if (j>2) {
4664                 printk(KERN_INFO
4665                         "md: resuming recovery of %s from checkpoint.\n",
4666                         mdname(mddev));
4667                 mddev->curr_resync = j;
4668         }
4669
4670         while (j < max_sectors) {
4671                 sector_t sectors;
4672
4673                 skipped = 0;
4674                 sectors = mddev->pers->sync_request(mddev, j, &skipped,
4675                                             currspeed < speed_min(mddev));
4676                 if (sectors == 0) {
4677                         set_bit(MD_RECOVERY_ERR, &mddev->recovery);
4678                         goto out;
4679                 }
4680
4681                 if (!skipped) { /* actual IO requested */
4682                         io_sectors += sectors;
4683                         atomic_add(sectors, &mddev->recovery_active);
4684                 }
4685
4686                 j += sectors;
4687                 if (j>1) mddev->curr_resync = j;
4688                 if (last_check == 0)
4689                         /* this is the earliers that rebuilt will be
4690                          * visible in /proc/mdstat
4691                          */
4692                         md_new_event(mddev);
4693
4694                 if (last_check + window > io_sectors || j == max_sectors)
4695                         continue;
4696
4697                 last_check = io_sectors;
4698
4699                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery) ||
4700                     test_bit(MD_RECOVERY_ERR, &mddev->recovery))
4701                         break;
4702
4703         repeat:
4704                 if (time_after_eq(jiffies, mark[last_mark] + SYNC_MARK_STEP )) {
4705                         /* step marks */
4706                         int next = (last_mark+1) % SYNC_MARKS;
4707
4708                         mddev->resync_mark = mark[next];
4709                         mddev->resync_mark_cnt = mark_cnt[next];
4710                         mark[next] = jiffies;
4711                         mark_cnt[next] = io_sectors - atomic_read(&mddev->recovery_active);
4712                         last_mark = next;
4713                 }
4714
4715
4716                 if (kthread_should_stop()) {
4717                         /*
4718                          * got a signal, exit.
4719                          */
4720                         printk(KERN_INFO
4721                                 "md: md_do_sync() got signal ... exiting\n");
4722                         set_bit(MD_RECOVERY_INTR, &mddev->recovery);
4723                         goto out;
4724                 }
4725
4726                 /*
4727                  * this loop exits only if either when we are slower than
4728                  * the 'hard' speed limit, or the system was IO-idle for
4729                  * a jiffy.
4730                  * the system might be non-idle CPU-wise, but we only care
4731                  * about not overloading the IO subsystem. (things like an
4732                  * e2fsck being done on the RAID array should execute fast)
4733                  */
4734                 mddev->queue->unplug_fn(mddev->queue);
4735                 cond_resched();
4736
4737                 currspeed = ((unsigned long)(io_sectors-mddev->resync_mark_cnt))/2
4738                         /((jiffies-mddev->resync_mark)/HZ +1) +1;
4739
4740                 if (currspeed > speed_min(mddev)) {
4741                         if ((currspeed > speed_max(mddev)) ||
4742                                         !is_mddev_idle(mddev)) {
4743                                 msleep(500);
4744                                 goto repeat;
4745                         }
4746                 }
4747         }
4748         printk(KERN_INFO "md: %s: sync done.\n",mdname(mddev));
4749         /*
4750          * this also signals 'finished resyncing' to md_stop
4751          */
4752  out:
4753         mddev->queue->unplug_fn(mddev->queue);
4754
4755         wait_event(mddev->recovery_wait, !atomic_read(&mddev->recovery_active));
4756
4757         /* tell personality that we are finished */
4758         mddev->pers->sync_request(mddev, max_sectors, &skipped, 1);
4759
4760         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4761             test_bit(MD_RECOVERY_SYNC, &mddev->recovery) &&
4762             !test_bit(MD_RECOVERY_CHECK, &mddev->recovery) &&
4763             mddev->curr_resync > 2 &&
4764             mddev->curr_resync >= mddev->recovery_cp) {
4765                 if (test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4766                         printk(KERN_INFO
4767                                 "md: checkpointing recovery of %s.\n",
4768                                 mdname(mddev));
4769                         mddev->recovery_cp = mddev->curr_resync;
4770                 } else
4771                         mddev->recovery_cp = MaxSector;
4772         }
4773
4774  skip:
4775         mddev->curr_resync = 0;
4776         wake_up(&resync_wait);
4777         set_bit(MD_RECOVERY_DONE, &mddev->recovery);
4778         md_wakeup_thread(mddev->thread);
4779 }
4780 EXPORT_SYMBOL_GPL(md_do_sync);
4781
4782
4783 /*
4784  * This routine is regularly called by all per-raid-array threads to
4785  * deal with generic issues like resync and super-block update.
4786  * Raid personalities that don't have a thread (linear/raid0) do not
4787  * need this as they never do any recovery or update the superblock.
4788  *
4789  * It does not do any resync itself, but rather "forks" off other threads
4790  * to do that as needed.
4791  * When it is determined that resync is needed, we set MD_RECOVERY_RUNNING in
4792  * "->recovery" and create a thread at ->sync_thread.
4793  * When the thread finishes it sets MD_RECOVERY_DONE (and might set MD_RECOVERY_ERR)
4794  * and wakeups up this thread which will reap the thread and finish up.
4795  * This thread also removes any faulty devices (with nr_pending == 0).
4796  *
4797  * The overall approach is:
4798  *  1/ if the superblock needs updating, update it.
4799  *  2/ If a recovery thread is running, don't do anything else.
4800  *  3/ If recovery has finished, clean up, possibly marking spares active.
4801  *  4/ If there are any faulty devices, remove them.
4802  *  5/ If array is degraded, try to add spares devices
4803  *  6/ If array has spares or is not in-sync, start a resync thread.
4804  */
4805 void md_check_recovery(mddev_t *mddev)
4806 {
4807         mdk_rdev_t *rdev;
4808         struct list_head *rtmp;
4809
4810
4811         if (mddev->bitmap)
4812                 bitmap_daemon_work(mddev->bitmap);
4813
4814         if (mddev->ro)
4815                 return;
4816
4817         if (signal_pending(current)) {
4818                 if (mddev->pers->sync_request) {
4819                         printk(KERN_INFO "md: %s in immediate safe mode\n",
4820                                mdname(mddev));
4821                         mddev->safemode = 2;
4822                 }
4823                 flush_signals(current);
4824         }
4825
4826         if ( ! (
4827                 mddev->sb_dirty ||
4828                 test_bit(MD_RECOVERY_NEEDED, &mddev->recovery) ||
4829                 test_bit(MD_RECOVERY_DONE, &mddev->recovery) ||
4830                 (mddev->safemode == 1) ||
4831                 (mddev->safemode == 2 && ! atomic_read(&mddev->writes_pending)
4832                  && !mddev->in_sync && mddev->recovery_cp == MaxSector)
4833                 ))
4834                 return;
4835
4836         if (mddev_trylock(mddev)==0) {
4837                 int spares =0;
4838
4839                 spin_lock_irq(&mddev->write_lock);
4840                 if (mddev->safemode && !atomic_read(&mddev->writes_pending) &&
4841                     !mddev->in_sync && mddev->recovery_cp == MaxSector) {
4842                         mddev->in_sync = 1;
4843                         mddev->sb_dirty = 1;
4844                 }
4845                 if (mddev->safemode == 1)
4846                         mddev->safemode = 0;
4847                 spin_unlock_irq(&mddev->write_lock);
4848
4849                 if (mddev->sb_dirty)
4850                         md_update_sb(mddev);
4851
4852
4853                 if (test_bit(MD_RECOVERY_RUNNING, &mddev->recovery) &&
4854                     !test_bit(MD_RECOVERY_DONE, &mddev->recovery)) {
4855                         /* resync/recovery still happening */
4856                         clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4857                         goto unlock;
4858                 }
4859                 if (mddev->sync_thread) {
4860                         /* resync has finished, collect result */
4861                         md_unregister_thread(mddev->sync_thread);
4862                         mddev->sync_thread = NULL;
4863                         if (!test_bit(MD_RECOVERY_ERR, &mddev->recovery) &&
4864                             !test_bit(MD_RECOVERY_INTR, &mddev->recovery)) {
4865                                 /* success...*/
4866                                 /* activate any spares */
4867                                 mddev->pers->spare_active(mddev);
4868                         }
4869                         md_update_sb(mddev);
4870
4871                         /* if array is no-longer degraded, then any saved_raid_disk
4872                          * information must be scrapped
4873                          */
4874                         if (!mddev->degraded)
4875                                 ITERATE_RDEV(mddev,rdev,rtmp)
4876                                         rdev->saved_raid_disk = -1;
4877
4878                         mddev->recovery = 0;
4879                         /* flag recovery needed just to double check */
4880                         set_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4881                         md_new_event(mddev);
4882                         goto unlock;
4883                 }
4884                 /* Clear some bits that don't mean anything, but
4885                  * might be left set
4886                  */
4887                 clear_bit(MD_RECOVERY_NEEDED, &mddev->recovery);
4888                 clear_bit(MD_RECOVERY_ERR, &mddev->recovery);
4889                 clear_bit(MD_RECOVERY_INTR, &mddev->recovery);
4890                 clear_bit(MD_RECOVERY_DONE, &mddev->recovery);
4891
4892                 /* no recovery is running.
4893                  * remove any failed drives, then
4894                  * add spares if possible.
4895                  * Spare are also removed and re-added, to allow
4896                  * the personality to fail the re-add.
4897                  */
4898                 ITERATE_RDEV(mddev,rdev,rtmp)
4899                         if (rdev->raid_disk >= 0 &&
4900                             (test_bit(Faulty, &rdev->flags) || ! test_bit(In_sync, &rdev->flags)) &&
4901                             atomic_read(&rdev->nr_pending)==0) {
4902                                 if (mddev->pers->hot_remove_disk(mddev, rdev->raid_disk)==0) {
4903                                         char nm[20];
4904                                         sprintf(nm,"rd%d", rdev->raid_disk);
4905                                         sysfs_remove_link(&mddev->kobj, nm);
4906                                         rdev->raid_disk = -1;
4907                                 }
4908                         }
4909
4910                 if (mddev->degraded) {
4911                         ITERATE_RDEV(mddev,rdev,rtmp)
4912                                 if (rdev->raid_disk < 0
4913                                     && !test_bit(Faulty, &rdev->flags)) {
4914                                         if (mddev->pers->hot_add_disk(mddev,rdev)) {
4915                                                 char nm[20];
4916                                                 sprintf(nm, "rd%d", rdev->raid_disk);
4917                                                 sysfs_create_link(&mddev->kobj, &rdev->kobj, nm);
4918                                                 spares++;
4919                                                 md_new_event(mddev);
4920                                         } else
4921                                                 break;
4922                                 }
4923                 }
4924
4925                 if (spares) {
4926                         clear_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4927                         clear_bit(MD_RECOVERY_CHECK, &mddev->recovery);
4928                 } else if (mddev->recovery_cp < MaxSector) {
4929                         set_bit(MD_RECOVERY_SYNC, &mddev->recovery);
4930                 } else if (!test_bit(MD_RECOVERY_SYNC, &mddev->recovery))
4931                         /* nothing to be done ... */
4932                         goto unlock;
4933
4934                 if (mddev->pers->sync_request) {
4935                         set_bit(MD_RECOVERY_RUNNING, &mddev->recovery);
4936                         if (spares && mddev->bitmap && ! mddev->bitmap->file) {
4937                                 /* We are adding a device or devices to an array
4938                                  * which has the bitmap stored on all devices.
4939                                  * So make sure all bitmap pages get written
4940                                  */
4941                                 bitmap_write_all(mddev->bitmap);
4942                         }
4943                         mddev->sync_thread = md_register_thread(md_do_sync,
4944                                                                 mddev,
4945                                                                 "%s_resync");
4946                         if (!mddev->sync_thread) {
4947                                 printk(KERN_ERR "%s: could not start resync"
4948                                         " thread...\n",
4949                                         mdname(mddev));
4950                                 /* leave the spares where they are, it shouldn't hurt */
4951                                 mddev->recovery = 0;
4952                         } else
4953                                 md_wakeup_thread(mddev->sync_thread);
4954                         md_new_event(mddev);
4955                 }
4956         unlock:
4957                 mddev_unlock(mddev);
4958         }
4959 }
4960
4961 static int md_notify_reboot(struct notifier_block *this,
4962                             unsigned long code, void *x)
4963 {
4964         struct list_head *tmp;
4965         mddev_t *mddev;
4966
4967         if ((code == SYS_DOWN) || (code == SYS_HALT) || (code == SYS_POWER_OFF)) {
4968
4969                 printk(KERN_INFO "md: stopping all md devices.\n");
4970
4971                 ITERATE_MDDEV(mddev,tmp)
4972                         if (mddev_trylock(mddev)==0)
4973                                 do_md_stop (mddev, 1);
4974                 /*
4975                  * certain more exotic SCSI devices are known to be
4976                  * volatile wrt too early system reboots. While the
4977                  * right place to handle this issue is the given
4978                  * driver, we do want to have a safe RAID driver ...
4979                  */
4980                 mdelay(1000*1);
4981         }
4982         return NOTIFY_DONE;
4983 }
4984
4985 static struct notifier_block md_notifier = {
4986         .notifier_call  = md_notify_reboot,
4987         .next           = NULL,
4988         .priority       = INT_MAX, /* before any real devices */
4989 };
4990
4991 static void md_geninit(void)
4992 {
4993         struct proc_dir_entry *p;
4994
4995         dprintk("md: sizeof(mdp_super_t) = %d\n", (int)sizeof(mdp_super_t));
4996
4997         p = create_proc_entry("mdstat", S_IRUGO, NULL);
4998         if (p)
4999                 p->proc_fops = &md_seq_fops;
5000 }
5001
5002 static int __init md_init(void)
5003 {
5004         int minor;
5005
5006         printk(KERN_INFO "md: md driver %d.%d.%d MAX_MD_DEVS=%d,"
5007                         " MD_SB_DISKS=%d\n",
5008                         MD_MAJOR_VERSION, MD_MINOR_VERSION,
5009                         MD_PATCHLEVEL_VERSION, MAX_MD_DEVS, MD_SB_DISKS);
5010         printk(KERN_INFO "md: bitmap version %d.%d\n", BITMAP_MAJOR_HI,
5011                         BITMAP_MINOR);
5012
5013         if (register_blkdev(MAJOR_NR, "md"))
5014                 return -1;
5015         if ((mdp_major=register_blkdev(0, "mdp"))<=0) {
5016                 unregister_blkdev(MAJOR_NR, "md");
5017                 return -1;
5018         }
5019         devfs_mk_dir("md");
5020         blk_register_region(MKDEV(MAJOR_NR, 0), MAX_MD_DEVS, THIS_MODULE,
5021                                 md_probe, NULL, NULL);
5022         blk_register_region(MKDEV(mdp_major, 0), MAX_MD_DEVS<<MdpMinorShift, THIS_MODULE,
5023                             md_probe, NULL, NULL);
5024
5025         for (minor=0; minor < MAX_MD_DEVS; ++minor)
5026                 devfs_mk_bdev(MKDEV(MAJOR_NR, minor),
5027                                 S_IFBLK|S_IRUSR|S_IWUSR,
5028                                 "md/%d", minor);
5029
5030         for (minor=0; minor < MAX_MD_DEVS; ++minor)
5031                 devfs_mk_bdev(MKDEV(mdp_major, minor<<MdpMinorShift),
5032                               S_IFBLK|S_IRUSR|S_IWUSR,
5033                               "md/mdp%d", minor);
5034
5035
5036         register_reboot_notifier(&md_notifier);
5037         raid_table_header = register_sysctl_table(raid_root_table, 1);
5038
5039         md_geninit();
5040         return (0);
5041 }
5042
5043
5044 #ifndef MODULE
5045
5046 /*
5047  * Searches all registered partitions for autorun RAID arrays
5048  * at boot time.
5049  */
5050 static dev_t detected_devices[128];
5051 static int dev_cnt;
5052
5053 void md_autodetect_dev(dev_t dev)
5054 {
5055         if (dev_cnt >= 0 && dev_cnt < 127)
5056                 detected_devices[dev_cnt++] = dev;
5057 }
5058
5059
5060 static void autostart_arrays(int part)
5061 {
5062         mdk_rdev_t *rdev;
5063         int i;
5064
5065         printk(KERN_INFO "md: Autodetecting RAID arrays.\n");
5066
5067         for (i = 0; i < dev_cnt; i++) {
5068                 dev_t dev = detected_devices[i];
5069
5070                 rdev = md_import_device(dev,0, 0);
5071                 if (IS_ERR(rdev))
5072                         continue;
5073
5074                 if (test_bit(Faulty, &rdev->flags)) {
5075                         MD_BUG();
5076                         continue;
5077                 }
5078                 list_add(&rdev->same_set, &pending_raid_disks);
5079         }
5080         dev_cnt = 0;
5081
5082         autorun_devices(part);
5083 }
5084
5085 #endif
5086
5087 static __exit void md_exit(void)
5088 {
5089         mddev_t *mddev;
5090         struct list_head *tmp;
5091         int i;
5092         blk_unregister_region(MKDEV(MAJOR_NR,0), MAX_MD_DEVS);
5093         blk_unregister_region(MKDEV(mdp_major,0), MAX_MD_DEVS << MdpMinorShift);
5094         for (i=0; i < MAX_MD_DEVS; i++)
5095                 devfs_remove("md/%d", i);
5096         for (i=0; i < MAX_MD_DEVS; i++)
5097                 devfs_remove("md/d%d", i);
5098
5099         devfs_remove("md");
5100
5101         unregister_blkdev(MAJOR_NR,"md");
5102         unregister_blkdev(mdp_major, "mdp");
5103         unregister_reboot_notifier(&md_notifier);
5104         unregister_sysctl_table(raid_table_header);
5105         remove_proc_entry("mdstat", NULL);
5106         ITERATE_MDDEV(mddev,tmp) {
5107                 struct gendisk *disk = mddev->gendisk;
5108                 if (!disk)
5109                         continue;
5110                 export_array(mddev);
5111                 del_gendisk(disk);
5112                 put_disk(disk);
5113                 mddev->gendisk = NULL;
5114                 mddev_put(mddev);
5115         }
5116 }
5117
5118 module_init(md_init)
5119 module_exit(md_exit)
5120
5121 static int get_ro(char *buffer, struct kernel_param *kp)
5122 {
5123         return sprintf(buffer, "%d", start_readonly);
5124 }
5125 static int set_ro(const char *val, struct kernel_param *kp)
5126 {
5127         char *e;
5128         int num = simple_strtoul(val, &e, 10);
5129         if (*val && (*e == '\0' || *e == '\n')) {
5130                 start_readonly = num;
5131                 return 0;
5132         }
5133         return -EINVAL;
5134 }
5135
5136 module_param_call(start_ro, set_ro, get_ro, NULL, 0600);
5137 module_param(start_dirty_degraded, int, 0644);
5138
5139
5140 EXPORT_SYMBOL(register_md_personality);
5141 EXPORT_SYMBOL(unregister_md_personality);
5142 EXPORT_SYMBOL(md_error);
5143 EXPORT_SYMBOL(md_done_sync);
5144 EXPORT_SYMBOL(md_write_start);
5145 EXPORT_SYMBOL(md_write_end);
5146 EXPORT_SYMBOL(md_register_thread);
5147 EXPORT_SYMBOL(md_unregister_thread);
5148 EXPORT_SYMBOL(md_wakeup_thread);
5149 EXPORT_SYMBOL(md_print_devices);
5150 EXPORT_SYMBOL(md_check_recovery);
5151 MODULE_LICENSE("GPL");
5152 MODULE_ALIAS("md");
5153 MODULE_ALIAS_BLOCKDEV_MAJOR(MD_MAJOR);